src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_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     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 104     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 105     __m256           dummy_mask,cutoff_mask;
 106     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 107     __m256           one     = _mm256_set1_ps(1.0);
 108     __m256           two     = _mm256_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm256_set1_ps(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 129     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 130     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 131     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139     j_coord_offsetE = 0;
 140     j_coord_offsetF = 0;
 141     j_coord_offsetG = 0;
 142     j_coord_offsetH = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm256_setzero_ps();
 171         fiy0             = _mm256_setzero_ps();
 172         fiz0             = _mm256_setzero_ps();
 173         fix1             = _mm256_setzero_ps();
 174         fiy1             = _mm256_setzero_ps();
 175         fiz1             = _mm256_setzero_ps();
 176         fix2             = _mm256_setzero_ps();
 177         fiy2             = _mm256_setzero_ps();
 178         fiz2             = _mm256_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm256_setzero_ps();
 182         vvdwsum          = _mm256_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             jnrE             = jjnr[jidx+4];
 194             jnrF             = jjnr[jidx+5];
 195             jnrG             = jjnr[jidx+6];
 196             jnrH             = jjnr[jidx+7];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201             j_coord_offsetE  = DIM*jnrE;
 202             j_coord_offsetF  = DIM*jnrF;
 203             j_coord_offsetG  = DIM*jnrG;
 204             j_coord_offsetH  = DIM*jnrH;
 205
 206             /* load j atom coordinates */
 207             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 209                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 210                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_ps(ix0,jx0);
 215             dy00             = _mm256_sub_ps(iy0,jy0);
 216             dz00             = _mm256_sub_ps(iz0,jz0);
 217             dx10             = _mm256_sub_ps(ix1,jx0);
 218             dy10             = _mm256_sub_ps(iy1,jy0);
 219             dz10             = _mm256_sub_ps(iz1,jz0);
 220             dx20             = _mm256_sub_ps(ix2,jx0);
 221             dy20             = _mm256_sub_ps(iy2,jy0);
 222             dz20             = _mm256_sub_ps(iz2,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 226             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 227             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 228
 229             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 230             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 231             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 232
 233             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 234             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 235             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 239                                                                  charge+jnrC+0,charge+jnrD+0,
 240                                                                  charge+jnrE+0,charge+jnrF+0,
 241                                                                  charge+jnrG+0,charge+jnrH+0);
 242             vdwjidx0A        = 2*vdwtype[jnrA+0];
 243             vdwjidx0B        = 2*vdwtype[jnrB+0];
 244             vdwjidx0C        = 2*vdwtype[jnrC+0];
 245             vdwjidx0D        = 2*vdwtype[jnrD+0];
 246             vdwjidx0E        = 2*vdwtype[jnrE+0];
 247             vdwjidx0F        = 2*vdwtype[jnrF+0];
 248             vdwjidx0G        = 2*vdwtype[jnrG+0];
 249             vdwjidx0H        = 2*vdwtype[jnrH+0];
 250
 251             fjx0             = _mm256_setzero_ps();
 252             fjy0             = _mm256_setzero_ps();
 253             fjz0             = _mm256_setzero_ps();
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             qq00             = _mm256_mul_ps(iq0,jq0);
 261             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 262                                             vdwioffsetptr0+vdwjidx0B,
 263                                             vdwioffsetptr0+vdwjidx0C,
 264                                             vdwioffsetptr0+vdwjidx0D,
 265                                             vdwioffsetptr0+vdwjidx0E,
 266                                             vdwioffsetptr0+vdwjidx0F,
 267                                             vdwioffsetptr0+vdwjidx0G,
 268                                             vdwioffsetptr0+vdwjidx0H,
 269                                             &c6_00,&c12_00);
 270
 271             /* COULOMB ELECTROSTATICS */
 272             velec            = _mm256_mul_ps(qq00,rinv00);
 273             felec            = _mm256_mul_ps(velec,rinvsq00);
 274
 275             /* LENNARD-JONES DISPERSION/REPULSION */
 276
 277             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 278             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 279             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 280             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 281             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velecsum         = _mm256_add_ps(velecsum,velec);
 285             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 286
 287             fscal            = _mm256_add_ps(felec,fvdw);
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm256_mul_ps(fscal,dx00);
 291             ty               = _mm256_mul_ps(fscal,dy00);
 292             tz               = _mm256_mul_ps(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm256_add_ps(fix0,tx);
 296             fiy0             = _mm256_add_ps(fiy0,ty);
 297             fiz0             = _mm256_add_ps(fiz0,tz);
 298
 299             fjx0             = _mm256_add_ps(fjx0,tx);
 300             fjy0             = _mm256_add_ps(fjy0,ty);
 301             fjz0             = _mm256_add_ps(fjz0,tz);
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq10             = _mm256_mul_ps(iq1,jq0);
 309
 310             /* COULOMB ELECTROSTATICS */
 311             velec            = _mm256_mul_ps(qq10,rinv10);
 312             felec            = _mm256_mul_ps(velec,rinvsq10);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velecsum         = _mm256_add_ps(velecsum,velec);
 316
 317             fscal            = felec;
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm256_mul_ps(fscal,dx10);
 321             ty               = _mm256_mul_ps(fscal,dy10);
 322             tz               = _mm256_mul_ps(fscal,dz10);
 323
 324             /* Update vectorial force */
 325             fix1             = _mm256_add_ps(fix1,tx);
 326             fiy1             = _mm256_add_ps(fiy1,ty);
 327             fiz1             = _mm256_add_ps(fiz1,tz);
 328
 329             fjx0             = _mm256_add_ps(fjx0,tx);
 330             fjy0             = _mm256_add_ps(fjy0,ty);
 331             fjz0             = _mm256_add_ps(fjz0,tz);
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq20             = _mm256_mul_ps(iq2,jq0);
 339
 340             /* COULOMB ELECTROSTATICS */
 341             velec            = _mm256_mul_ps(qq20,rinv20);
 342             felec            = _mm256_mul_ps(velec,rinvsq20);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm256_add_ps(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_ps(fscal,dx20);
 351             ty               = _mm256_mul_ps(fscal,dy20);
 352             tz               = _mm256_mul_ps(fscal,dz20);
 353
 354             /* Update vectorial force */
 355             fix2             = _mm256_add_ps(fix2,tx);
 356             fiy2             = _mm256_add_ps(fiy2,ty);
 357             fiz2             = _mm256_add_ps(fiz2,tz);
 358
 359             fjx0             = _mm256_add_ps(fjx0,tx);
 360             fjy0             = _mm256_add_ps(fjy0,ty);
 361             fjz0             = _mm256_add_ps(fjz0,tz);
 362
 363             fjptrA             = f+j_coord_offsetA;
 364             fjptrB             = f+j_coord_offsetB;
 365             fjptrC             = f+j_coord_offsetC;
 366             fjptrD             = f+j_coord_offsetD;
 367             fjptrE             = f+j_coord_offsetE;
 368             fjptrF             = f+j_coord_offsetF;
 369             fjptrG             = f+j_coord_offsetG;
 370             fjptrH             = f+j_coord_offsetH;
 371
 372             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 373
 374             /* Inner loop uses 96 flops */
 375         }
 376
 377         if(jidx<j_index_end)
 378         {
 379
 380             /* Get j neighbor index, and coordinate index */
 381             jnrlistA         = jjnr[jidx];
 382             jnrlistB         = jjnr[jidx+1];
 383             jnrlistC         = jjnr[jidx+2];
 384             jnrlistD         = jjnr[jidx+3];
 385             jnrlistE         = jjnr[jidx+4];
 386             jnrlistF         = jjnr[jidx+5];
 387             jnrlistG         = jjnr[jidx+6];
 388             jnrlistH         = jjnr[jidx+7];
 389             /* Sign of each element will be negative for non-real atoms.
 390              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 391              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 392              */
 393             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 394                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 395
 396             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 397             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 398             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 399             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 400             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 401             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 402             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 403             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 404             j_coord_offsetA  = DIM*jnrA;
 405             j_coord_offsetB  = DIM*jnrB;
 406             j_coord_offsetC  = DIM*jnrC;
 407             j_coord_offsetD  = DIM*jnrD;
 408             j_coord_offsetE  = DIM*jnrE;
 409             j_coord_offsetF  = DIM*jnrF;
 410             j_coord_offsetG  = DIM*jnrG;
 411             j_coord_offsetH  = DIM*jnrH;
 412
 413             /* load j atom coordinates */
 414             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 415                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 416                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 417                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 418                                                  &jx0,&jy0,&jz0);
 419
 420             /* Calculate displacement vector */
 421             dx00             = _mm256_sub_ps(ix0,jx0);
 422             dy00             = _mm256_sub_ps(iy0,jy0);
 423             dz00             = _mm256_sub_ps(iz0,jz0);
 424             dx10             = _mm256_sub_ps(ix1,jx0);
 425             dy10             = _mm256_sub_ps(iy1,jy0);
 426             dz10             = _mm256_sub_ps(iz1,jz0);
 427             dx20             = _mm256_sub_ps(ix2,jx0);
 428             dy20             = _mm256_sub_ps(iy2,jy0);
 429             dz20             = _mm256_sub_ps(iz2,jz0);
 430
 431             /* Calculate squared distance and things based on it */
 432             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 433             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 434             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 435
 436             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 437             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 438             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 439
 440             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 441             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 442             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 443
 444             /* Load parameters for j particles */
 445             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 446                                                                  charge+jnrC+0,charge+jnrD+0,
 447                                                                  charge+jnrE+0,charge+jnrF+0,
 448                                                                  charge+jnrG+0,charge+jnrH+0);
 449             vdwjidx0A        = 2*vdwtype[jnrA+0];
 450             vdwjidx0B        = 2*vdwtype[jnrB+0];
 451             vdwjidx0C        = 2*vdwtype[jnrC+0];
 452             vdwjidx0D        = 2*vdwtype[jnrD+0];
 453             vdwjidx0E        = 2*vdwtype[jnrE+0];
 454             vdwjidx0F        = 2*vdwtype[jnrF+0];
 455             vdwjidx0G        = 2*vdwtype[jnrG+0];
 456             vdwjidx0H        = 2*vdwtype[jnrH+0];
 457
 458             fjx0             = _mm256_setzero_ps();
 459             fjy0             = _mm256_setzero_ps();
 460             fjz0             = _mm256_setzero_ps();
 461
 462             /**************************
 463              * CALCULATE INTERACTIONS *
 464              **************************/
 465
 466             /* Compute parameters for interactions between i and j atoms */
 467             qq00             = _mm256_mul_ps(iq0,jq0);
 468             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 469                                             vdwioffsetptr0+vdwjidx0B,
 470                                             vdwioffsetptr0+vdwjidx0C,
 471                                             vdwioffsetptr0+vdwjidx0D,
 472                                             vdwioffsetptr0+vdwjidx0E,
 473                                             vdwioffsetptr0+vdwjidx0F,
 474                                             vdwioffsetptr0+vdwjidx0G,
 475                                             vdwioffsetptr0+vdwjidx0H,
 476                                             &c6_00,&c12_00);
 477
 478             /* COULOMB ELECTROSTATICS */
 479             velec            = _mm256_mul_ps(qq00,rinv00);
 480             felec            = _mm256_mul_ps(velec,rinvsq00);
 481
 482             /* LENNARD-JONES DISPERSION/REPULSION */
 483
 484             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 485             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 486             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 487             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 488             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 489
 490             /* Update potential sum for this i atom from the interaction with this j atom. */
 491             velec            = _mm256_andnot_ps(dummy_mask,velec);
 492             velecsum         = _mm256_add_ps(velecsum,velec);
 493             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 494             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 495
 496             fscal            = _mm256_add_ps(felec,fvdw);
 497
 498             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 499
 500             /* Calculate temporary vectorial force */
 501             tx               = _mm256_mul_ps(fscal,dx00);
 502             ty               = _mm256_mul_ps(fscal,dy00);
 503             tz               = _mm256_mul_ps(fscal,dz00);
 504
 505             /* Update vectorial force */
 506             fix0             = _mm256_add_ps(fix0,tx);
 507             fiy0             = _mm256_add_ps(fiy0,ty);
 508             fiz0             = _mm256_add_ps(fiz0,tz);
 509
 510             fjx0             = _mm256_add_ps(fjx0,tx);
 511             fjy0             = _mm256_add_ps(fjy0,ty);
 512             fjz0             = _mm256_add_ps(fjz0,tz);
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             /* Compute parameters for interactions between i and j atoms */
 519             qq10             = _mm256_mul_ps(iq1,jq0);
 520
 521             /* COULOMB ELECTROSTATICS */
 522             velec            = _mm256_mul_ps(qq10,rinv10);
 523             felec            = _mm256_mul_ps(velec,rinvsq10);
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             velec            = _mm256_andnot_ps(dummy_mask,velec);
 527             velecsum         = _mm256_add_ps(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 532
 533             /* Calculate temporary vectorial force */
 534             tx               = _mm256_mul_ps(fscal,dx10);
 535             ty               = _mm256_mul_ps(fscal,dy10);
 536             tz               = _mm256_mul_ps(fscal,dz10);
 537
 538             /* Update vectorial force */
 539             fix1             = _mm256_add_ps(fix1,tx);
 540             fiy1             = _mm256_add_ps(fiy1,ty);
 541             fiz1             = _mm256_add_ps(fiz1,tz);
 542
 543             fjx0             = _mm256_add_ps(fjx0,tx);
 544             fjy0             = _mm256_add_ps(fjy0,ty);
 545             fjz0             = _mm256_add_ps(fjz0,tz);
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             /* Compute parameters for interactions between i and j atoms */
 552             qq20             = _mm256_mul_ps(iq2,jq0);
 553
 554             /* COULOMB ELECTROSTATICS */
 555             velec            = _mm256_mul_ps(qq20,rinv20);
 556             felec            = _mm256_mul_ps(velec,rinvsq20);
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             velec            = _mm256_andnot_ps(dummy_mask,velec);
 560             velecsum         = _mm256_add_ps(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm256_mul_ps(fscal,dx20);
 568             ty               = _mm256_mul_ps(fscal,dy20);
 569             tz               = _mm256_mul_ps(fscal,dz20);
 570
 571             /* Update vectorial force */
 572             fix2             = _mm256_add_ps(fix2,tx);
 573             fiy2             = _mm256_add_ps(fiy2,ty);
 574             fiz2             = _mm256_add_ps(fiz2,tz);
 575
 576             fjx0             = _mm256_add_ps(fjx0,tx);
 577             fjy0             = _mm256_add_ps(fjy0,ty);
 578             fjz0             = _mm256_add_ps(fjz0,tz);
 579
 580             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 581             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 582             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 583             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 584             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 585             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 586             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 587             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 588
 589             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 590
 591             /* Inner loop uses 96 flops */
 592         }
 593
 594         /* End of innermost loop */
 595
 596         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 597                                                  f+i_coord_offset,fshift+i_shift_offset);
 598
 599         ggid                        = gid[iidx];
 600         /* Update potential energies */
 601         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 602         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 603
 604         /* Increment number of inner iterations */
 605         inneriter                  += j_index_end - j_index_start;
 606
 607         /* Outer loop uses 20 flops */
 608     }
 609
 610     /* Increment number of outer iterations */
 611     outeriter        += nri;
 612
 613     /* Update outer/inner flops */
 614
 615     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
 616 }
 617 /*
 618  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_single
 619  * Electrostatics interaction: Coulomb
 620  * VdW interaction:            LennardJones
 621  * Geometry:                   Water3-Particle
 622  * Calculate force/pot:        Force
 623  */
 624 void
 625 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_single
 626                     (t_nblist                    * gmx_restrict       nlist,
 627                      rvec                        * gmx_restrict          xx,
 628                      rvec                        * gmx_restrict          ff,
 629                      t_forcerec                  * gmx_restrict          fr,
 630                      t_mdatoms                   * gmx_restrict     mdatoms,
 631                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 632                      t_nrnb                      * gmx_restrict        nrnb)
 633 {
 634     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 635      * just 0 for non-waters.
 636      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 637      * jnr indices corresponding to data put in the four positions in the SIMD register.
 638      */
 639     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 640     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 641     int              jnrA,jnrB,jnrC,jnrD;
 642     int              jnrE,jnrF,jnrG,jnrH;
 643     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 644     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 645     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 646     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 647     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 648     real             rcutoff_scalar;
 649     real             *shiftvec,*fshift,*x,*f;
 650     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 651     real             scratch[4*DIM];
 652     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 653     real *           vdwioffsetptr0;
 654     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 655     real *           vdwioffsetptr1;
 656     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 657     real *           vdwioffsetptr2;
 658     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 659     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 660     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 661     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 662     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 663     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 664     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 665     real             *charge;
 666     int              nvdwtype;
 667     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 668     int              *vdwtype;
 669     real             *vdwparam;
 670     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 671     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 672     __m256           dummy_mask,cutoff_mask;
 673     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 674     __m256           one     = _mm256_set1_ps(1.0);
 675     __m256           two     = _mm256_set1_ps(2.0);
 676     x                = xx[0];
 677     f                = ff[0];
 678
 679     nri              = nlist->nri;
 680     iinr             = nlist->iinr;
 681     jindex           = nlist->jindex;
 682     jjnr             = nlist->jjnr;
 683     shiftidx         = nlist->shift;
 684     gid              = nlist->gid;
 685     shiftvec         = fr->shift_vec[0];
 686     fshift           = fr->fshift[0];
 687     facel            = _mm256_set1_ps(fr->epsfac);
 688     charge           = mdatoms->chargeA;
 689     nvdwtype         = fr->ntype;
 690     vdwparam         = fr->nbfp;
 691     vdwtype          = mdatoms->typeA;
 692
 693     /* Setup water-specific parameters */
 694     inr              = nlist->iinr[0];
 695     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 696     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 697     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 698     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 699
 700     /* Avoid stupid compiler warnings */
 701     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 702     j_coord_offsetA = 0;
 703     j_coord_offsetB = 0;
 704     j_coord_offsetC = 0;
 705     j_coord_offsetD = 0;
 706     j_coord_offsetE = 0;
 707     j_coord_offsetF = 0;
 708     j_coord_offsetG = 0;
 709     j_coord_offsetH = 0;
 710
 711     outeriter        = 0;
 712     inneriter        = 0;
 713
 714     for(iidx=0;iidx<4*DIM;iidx++)
 715     {
 716         scratch[iidx] = 0.0;
 717     }
 718
 719     /* Start outer loop over neighborlists */
 720     for(iidx=0; iidx<nri; iidx++)
 721     {
 722         /* Load shift vector for this list */
 723         i_shift_offset   = DIM*shiftidx[iidx];
 724
 725         /* Load limits for loop over neighbors */
 726         j_index_start    = jindex[iidx];
 727         j_index_end      = jindex[iidx+1];
 728
 729         /* Get outer coordinate index */
 730         inr              = iinr[iidx];
 731         i_coord_offset   = DIM*inr;
 732
 733         /* Load i particle coords and add shift vector */
 734         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 735                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 736
 737         fix0             = _mm256_setzero_ps();
 738         fiy0             = _mm256_setzero_ps();
 739         fiz0             = _mm256_setzero_ps();
 740         fix1             = _mm256_setzero_ps();
 741         fiy1             = _mm256_setzero_ps();
 742         fiz1             = _mm256_setzero_ps();
 743         fix2             = _mm256_setzero_ps();
 744         fiy2             = _mm256_setzero_ps();
 745         fiz2             = _mm256_setzero_ps();
 746
 747         /* Start inner kernel loop */
 748         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 749         {
 750
 751             /* Get j neighbor index, and coordinate index */
 752             jnrA             = jjnr[jidx];
 753             jnrB             = jjnr[jidx+1];
 754             jnrC             = jjnr[jidx+2];
 755             jnrD             = jjnr[jidx+3];
 756             jnrE             = jjnr[jidx+4];
 757             jnrF             = jjnr[jidx+5];
 758             jnrG             = jjnr[jidx+6];
 759             jnrH             = jjnr[jidx+7];
 760             j_coord_offsetA  = DIM*jnrA;
 761             j_coord_offsetB  = DIM*jnrB;
 762             j_coord_offsetC  = DIM*jnrC;
 763             j_coord_offsetD  = DIM*jnrD;
 764             j_coord_offsetE  = DIM*jnrE;
 765             j_coord_offsetF  = DIM*jnrF;
 766             j_coord_offsetG  = DIM*jnrG;
 767             j_coord_offsetH  = DIM*jnrH;
 768
 769             /* load j atom coordinates */
 770             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 771                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 772                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 773                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 774                                                  &jx0,&jy0,&jz0);
 775
 776             /* Calculate displacement vector */
 777             dx00             = _mm256_sub_ps(ix0,jx0);
 778             dy00             = _mm256_sub_ps(iy0,jy0);
 779             dz00             = _mm256_sub_ps(iz0,jz0);
 780             dx10             = _mm256_sub_ps(ix1,jx0);
 781             dy10             = _mm256_sub_ps(iy1,jy0);
 782             dz10             = _mm256_sub_ps(iz1,jz0);
 783             dx20             = _mm256_sub_ps(ix2,jx0);
 784             dy20             = _mm256_sub_ps(iy2,jy0);
 785             dz20             = _mm256_sub_ps(iz2,jz0);
 786
 787             /* Calculate squared distance and things based on it */
 788             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 789             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 790             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 791
 792             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 793             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 794             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 795
 796             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 797             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 798             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 799
 800             /* Load parameters for j particles */
 801             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 802                                                                  charge+jnrC+0,charge+jnrD+0,
 803                                                                  charge+jnrE+0,charge+jnrF+0,
 804                                                                  charge+jnrG+0,charge+jnrH+0);
 805             vdwjidx0A        = 2*vdwtype[jnrA+0];
 806             vdwjidx0B        = 2*vdwtype[jnrB+0];
 807             vdwjidx0C        = 2*vdwtype[jnrC+0];
 808             vdwjidx0D        = 2*vdwtype[jnrD+0];
 809             vdwjidx0E        = 2*vdwtype[jnrE+0];
 810             vdwjidx0F        = 2*vdwtype[jnrF+0];
 811             vdwjidx0G        = 2*vdwtype[jnrG+0];
 812             vdwjidx0H        = 2*vdwtype[jnrH+0];
 813
 814             fjx0             = _mm256_setzero_ps();
 815             fjy0             = _mm256_setzero_ps();
 816             fjz0             = _mm256_setzero_ps();
 817
 818             /**************************
 819              * CALCULATE INTERACTIONS *
 820              **************************/
 821
 822             /* Compute parameters for interactions between i and j atoms */
 823             qq00             = _mm256_mul_ps(iq0,jq0);
 824             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 825                                             vdwioffsetptr0+vdwjidx0B,
 826                                             vdwioffsetptr0+vdwjidx0C,
 827                                             vdwioffsetptr0+vdwjidx0D,
 828                                             vdwioffsetptr0+vdwjidx0E,
 829                                             vdwioffsetptr0+vdwjidx0F,
 830                                             vdwioffsetptr0+vdwjidx0G,
 831                                             vdwioffsetptr0+vdwjidx0H,
 832                                             &c6_00,&c12_00);
 833
 834             /* COULOMB ELECTROSTATICS */
 835             velec            = _mm256_mul_ps(qq00,rinv00);
 836             felec            = _mm256_mul_ps(velec,rinvsq00);
 837
 838             /* LENNARD-JONES DISPERSION/REPULSION */
 839
 840             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 841             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 842
 843             fscal            = _mm256_add_ps(felec,fvdw);
 844
 845             /* Calculate temporary vectorial force */
 846             tx               = _mm256_mul_ps(fscal,dx00);
 847             ty               = _mm256_mul_ps(fscal,dy00);
 848             tz               = _mm256_mul_ps(fscal,dz00);
 849
 850             /* Update vectorial force */
 851             fix0             = _mm256_add_ps(fix0,tx);
 852             fiy0             = _mm256_add_ps(fiy0,ty);
 853             fiz0             = _mm256_add_ps(fiz0,tz);
 854
 855             fjx0             = _mm256_add_ps(fjx0,tx);
 856             fjy0             = _mm256_add_ps(fjy0,ty);
 857             fjz0             = _mm256_add_ps(fjz0,tz);
 858
 859             /**************************
 860              * CALCULATE INTERACTIONS *
 861              **************************/
 862
 863             /* Compute parameters for interactions between i and j atoms */
 864             qq10             = _mm256_mul_ps(iq1,jq0);
 865
 866             /* COULOMB ELECTROSTATICS */
 867             velec            = _mm256_mul_ps(qq10,rinv10);
 868             felec            = _mm256_mul_ps(velec,rinvsq10);
 869
 870             fscal            = felec;
 871
 872             /* Calculate temporary vectorial force */
 873             tx               = _mm256_mul_ps(fscal,dx10);
 874             ty               = _mm256_mul_ps(fscal,dy10);
 875             tz               = _mm256_mul_ps(fscal,dz10);
 876
 877             /* Update vectorial force */
 878             fix1             = _mm256_add_ps(fix1,tx);
 879             fiy1             = _mm256_add_ps(fiy1,ty);
 880             fiz1             = _mm256_add_ps(fiz1,tz);
 881
 882             fjx0             = _mm256_add_ps(fjx0,tx);
 883             fjy0             = _mm256_add_ps(fjy0,ty);
 884             fjz0             = _mm256_add_ps(fjz0,tz);
 885
 886             /**************************
 887              * CALCULATE INTERACTIONS *
 888              **************************/
 889
 890             /* Compute parameters for interactions between i and j atoms */
 891             qq20             = _mm256_mul_ps(iq2,jq0);
 892
 893             /* COULOMB ELECTROSTATICS */
 894             velec            = _mm256_mul_ps(qq20,rinv20);
 895             felec            = _mm256_mul_ps(velec,rinvsq20);
 896
 897             fscal            = felec;
 898
 899             /* Calculate temporary vectorial force */
 900             tx               = _mm256_mul_ps(fscal,dx20);
 901             ty               = _mm256_mul_ps(fscal,dy20);
 902             tz               = _mm256_mul_ps(fscal,dz20);
 903
 904             /* Update vectorial force */
 905             fix2             = _mm256_add_ps(fix2,tx);
 906             fiy2             = _mm256_add_ps(fiy2,ty);
 907             fiz2             = _mm256_add_ps(fiz2,tz);
 908
 909             fjx0             = _mm256_add_ps(fjx0,tx);
 910             fjy0             = _mm256_add_ps(fjy0,ty);
 911             fjz0             = _mm256_add_ps(fjz0,tz);
 912
 913             fjptrA             = f+j_coord_offsetA;
 914             fjptrB             = f+j_coord_offsetB;
 915             fjptrC             = f+j_coord_offsetC;
 916             fjptrD             = f+j_coord_offsetD;
 917             fjptrE             = f+j_coord_offsetE;
 918             fjptrF             = f+j_coord_offsetF;
 919             fjptrG             = f+j_coord_offsetG;
 920             fjptrH             = f+j_coord_offsetH;
 921
 922             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 923
 924             /* Inner loop uses 88 flops */
 925         }
 926
 927         if(jidx<j_index_end)
 928         {
 929
 930             /* Get j neighbor index, and coordinate index */
 931             jnrlistA         = jjnr[jidx];
 932             jnrlistB         = jjnr[jidx+1];
 933             jnrlistC         = jjnr[jidx+2];
 934             jnrlistD         = jjnr[jidx+3];
 935             jnrlistE         = jjnr[jidx+4];
 936             jnrlistF         = jjnr[jidx+5];
 937             jnrlistG         = jjnr[jidx+6];
 938             jnrlistH         = jjnr[jidx+7];
 939             /* Sign of each element will be negative for non-real atoms.
 940              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 941              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 942              */
 943             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 944                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 945
 946             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 947             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 948             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 949             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 950             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 951             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 952             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 953             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 954             j_coord_offsetA  = DIM*jnrA;
 955             j_coord_offsetB  = DIM*jnrB;
 956             j_coord_offsetC  = DIM*jnrC;
 957             j_coord_offsetD  = DIM*jnrD;
 958             j_coord_offsetE  = DIM*jnrE;
 959             j_coord_offsetF  = DIM*jnrF;
 960             j_coord_offsetG  = DIM*jnrG;
 961             j_coord_offsetH  = DIM*jnrH;
 962
 963             /* load j atom coordinates */
 964             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 965                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 966                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 967                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 968                                                  &jx0,&jy0,&jz0);
 969
 970             /* Calculate displacement vector */
 971             dx00             = _mm256_sub_ps(ix0,jx0);
 972             dy00             = _mm256_sub_ps(iy0,jy0);
 973             dz00             = _mm256_sub_ps(iz0,jz0);
 974             dx10             = _mm256_sub_ps(ix1,jx0);
 975             dy10             = _mm256_sub_ps(iy1,jy0);
 976             dz10             = _mm256_sub_ps(iz1,jz0);
 977             dx20             = _mm256_sub_ps(ix2,jx0);
 978             dy20             = _mm256_sub_ps(iy2,jy0);
 979             dz20             = _mm256_sub_ps(iz2,jz0);
 980
 981             /* Calculate squared distance and things based on it */
 982             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 983             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 984             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 985
 986             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 987             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 988             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 989
 990             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 991             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 992             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 993
 994             /* Load parameters for j particles */
 995             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 996                                                                  charge+jnrC+0,charge+jnrD+0,
 997                                                                  charge+jnrE+0,charge+jnrF+0,
 998                                                                  charge+jnrG+0,charge+jnrH+0);
 999             vdwjidx0A        = 2*vdwtype[jnrA+0];
1000             vdwjidx0B        = 2*vdwtype[jnrB+0];
1001             vdwjidx0C        = 2*vdwtype[jnrC+0];
1002             vdwjidx0D        = 2*vdwtype[jnrD+0];
1003             vdwjidx0E        = 2*vdwtype[jnrE+0];
1004             vdwjidx0F        = 2*vdwtype[jnrF+0];
1005             vdwjidx0G        = 2*vdwtype[jnrG+0];
1006             vdwjidx0H        = 2*vdwtype[jnrH+0];
1007
1008             fjx0             = _mm256_setzero_ps();
1009             fjy0             = _mm256_setzero_ps();
1010             fjz0             = _mm256_setzero_ps();
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             /* Compute parameters for interactions between i and j atoms */
1017             qq00             = _mm256_mul_ps(iq0,jq0);
1018             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1019                                             vdwioffsetptr0+vdwjidx0B,
1020                                             vdwioffsetptr0+vdwjidx0C,
1021                                             vdwioffsetptr0+vdwjidx0D,
1022                                             vdwioffsetptr0+vdwjidx0E,
1023                                             vdwioffsetptr0+vdwjidx0F,
1024                                             vdwioffsetptr0+vdwjidx0G,
1025                                             vdwioffsetptr0+vdwjidx0H,
1026                                             &c6_00,&c12_00);
1027
1028             /* COULOMB ELECTROSTATICS */
1029             velec            = _mm256_mul_ps(qq00,rinv00);
1030             felec            = _mm256_mul_ps(velec,rinvsq00);
1031
1032             /* LENNARD-JONES DISPERSION/REPULSION */
1033
1034             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1035             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1036
1037             fscal            = _mm256_add_ps(felec,fvdw);
1038
1039             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1040
1041             /* Calculate temporary vectorial force */
1042             tx               = _mm256_mul_ps(fscal,dx00);
1043             ty               = _mm256_mul_ps(fscal,dy00);
1044             tz               = _mm256_mul_ps(fscal,dz00);
1045
1046             /* Update vectorial force */
1047             fix0             = _mm256_add_ps(fix0,tx);
1048             fiy0             = _mm256_add_ps(fiy0,ty);
1049             fiz0             = _mm256_add_ps(fiz0,tz);
1050
1051             fjx0             = _mm256_add_ps(fjx0,tx);
1052             fjy0             = _mm256_add_ps(fjy0,ty);
1053             fjz0             = _mm256_add_ps(fjz0,tz);
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             /* Compute parameters for interactions between i and j atoms */
1060             qq10             = _mm256_mul_ps(iq1,jq0);
1061
1062             /* COULOMB ELECTROSTATICS */
1063             velec            = _mm256_mul_ps(qq10,rinv10);
1064             felec            = _mm256_mul_ps(velec,rinvsq10);
1065
1066             fscal            = felec;
1067
1068             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1069
1070             /* Calculate temporary vectorial force */
1071             tx               = _mm256_mul_ps(fscal,dx10);
1072             ty               = _mm256_mul_ps(fscal,dy10);
1073             tz               = _mm256_mul_ps(fscal,dz10);
1074
1075             /* Update vectorial force */
1076             fix1             = _mm256_add_ps(fix1,tx);
1077             fiy1             = _mm256_add_ps(fiy1,ty);
1078             fiz1             = _mm256_add_ps(fiz1,tz);
1079
1080             fjx0             = _mm256_add_ps(fjx0,tx);
1081             fjy0             = _mm256_add_ps(fjy0,ty);
1082             fjz0             = _mm256_add_ps(fjz0,tz);
1083
1084             /**************************
1085              * CALCULATE INTERACTIONS *
1086              **************************/
1087
1088             /* Compute parameters for interactions between i and j atoms */
1089             qq20             = _mm256_mul_ps(iq2,jq0);
1090
1091             /* COULOMB ELECTROSTATICS */
1092             velec            = _mm256_mul_ps(qq20,rinv20);
1093             felec            = _mm256_mul_ps(velec,rinvsq20);
1094
1095             fscal            = felec;
1096
1097             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1098
1099             /* Calculate temporary vectorial force */
1100             tx               = _mm256_mul_ps(fscal,dx20);
1101             ty               = _mm256_mul_ps(fscal,dy20);
1102             tz               = _mm256_mul_ps(fscal,dz20);
1103
1104             /* Update vectorial force */
1105             fix2             = _mm256_add_ps(fix2,tx);
1106             fiy2             = _mm256_add_ps(fiy2,ty);
1107             fiz2             = _mm256_add_ps(fiz2,tz);
1108
1109             fjx0             = _mm256_add_ps(fjx0,tx);
1110             fjy0             = _mm256_add_ps(fjy0,ty);
1111             fjz0             = _mm256_add_ps(fjz0,tz);
1112
1113             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1114             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1115             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1116             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1117             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1118             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1119             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1120             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1121
1122             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1123
1124             /* Inner loop uses 88 flops */
1125         }
1126
1127         /* End of innermost loop */
1128
1129         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1130                                                  f+i_coord_offset,fshift+i_shift_offset);
1131
1132         /* Increment number of inner iterations */
1133         inneriter                  += j_index_end - j_index_start;
1134
1135         /* Outer loop uses 18 flops */
1136     }
1137
1138     /* Increment number of outer iterations */
1139     outeriter        += nri;
1140
1141     /* Update outer/inner flops */
1142
1143     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);
1144 }