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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256           dummy_mask,cutoff_mask;
 108     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 109     __m256           one     = _mm256_set1_ps(1.0);
 110     __m256           two     = _mm256_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 131     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 132     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 133     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141     j_coord_offsetE = 0;
 142     j_coord_offsetF = 0;
 143     j_coord_offsetG = 0;
 144     j_coord_offsetH = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm256_setzero_ps();
 173         fiy0             = _mm256_setzero_ps();
 174         fiz0             = _mm256_setzero_ps();
 175         fix1             = _mm256_setzero_ps();
 176         fiy1             = _mm256_setzero_ps();
 177         fiz1             = _mm256_setzero_ps();
 178         fix2             = _mm256_setzero_ps();
 179         fiy2             = _mm256_setzero_ps();
 180         fiz2             = _mm256_setzero_ps();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm256_setzero_ps();
 184         vvdwsum          = _mm256_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             jnrE             = jjnr[jidx+4];
 196             jnrF             = jjnr[jidx+5];
 197             jnrG             = jjnr[jidx+6];
 198             jnrH             = jjnr[jidx+7];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203             j_coord_offsetE  = DIM*jnrE;
 204             j_coord_offsetF  = DIM*jnrF;
 205             j_coord_offsetG  = DIM*jnrG;
 206             j_coord_offsetH  = DIM*jnrH;
 207
 208             /* load j atom coordinates */
 209             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 211                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 212                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 213                                                  &jx0,&jy0,&jz0);
 214
 215             /* Calculate displacement vector */
 216             dx00             = _mm256_sub_ps(ix0,jx0);
 217             dy00             = _mm256_sub_ps(iy0,jy0);
 218             dz00             = _mm256_sub_ps(iz0,jz0);
 219             dx10             = _mm256_sub_ps(ix1,jx0);
 220             dy10             = _mm256_sub_ps(iy1,jy0);
 221             dz10             = _mm256_sub_ps(iz1,jz0);
 222             dx20             = _mm256_sub_ps(ix2,jx0);
 223             dy20             = _mm256_sub_ps(iy2,jy0);
 224             dz20             = _mm256_sub_ps(iz2,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 228             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 229             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 230
 231             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 232             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 233             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 234
 235             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 236             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 237             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 241                                                                  charge+jnrC+0,charge+jnrD+0,
 242                                                                  charge+jnrE+0,charge+jnrF+0,
 243                                                                  charge+jnrG+0,charge+jnrH+0);
 244             vdwjidx0A        = 2*vdwtype[jnrA+0];
 245             vdwjidx0B        = 2*vdwtype[jnrB+0];
 246             vdwjidx0C        = 2*vdwtype[jnrC+0];
 247             vdwjidx0D        = 2*vdwtype[jnrD+0];
 248             vdwjidx0E        = 2*vdwtype[jnrE+0];
 249             vdwjidx0F        = 2*vdwtype[jnrF+0];
 250             vdwjidx0G        = 2*vdwtype[jnrG+0];
 251             vdwjidx0H        = 2*vdwtype[jnrH+0];
 252
 253             fjx0             = _mm256_setzero_ps();
 254             fjy0             = _mm256_setzero_ps();
 255             fjz0             = _mm256_setzero_ps();
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             qq00             = _mm256_mul_ps(iq0,jq0);
 263             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 264                                             vdwioffsetptr0+vdwjidx0B,
 265                                             vdwioffsetptr0+vdwjidx0C,
 266                                             vdwioffsetptr0+vdwjidx0D,
 267                                             vdwioffsetptr0+vdwjidx0E,
 268                                             vdwioffsetptr0+vdwjidx0F,
 269                                             vdwioffsetptr0+vdwjidx0G,
 270                                             vdwioffsetptr0+vdwjidx0H,
 271                                             &c6_00,&c12_00);
 272
 273             /* COULOMB ELECTROSTATICS */
 274             velec            = _mm256_mul_ps(qq00,rinv00);
 275             felec            = _mm256_mul_ps(velec,rinvsq00);
 276
 277             /* LENNARD-JONES DISPERSION/REPULSION */
 278
 279             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 280             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 281             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 282             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 283             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 284
 285             /* Update potential sum for this i atom from the interaction with this j atom. */
 286             velecsum         = _mm256_add_ps(velecsum,velec);
 287             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 288
 289             fscal            = _mm256_add_ps(felec,fvdw);
 290
 291             /* Calculate temporary vectorial force */
 292             tx               = _mm256_mul_ps(fscal,dx00);
 293             ty               = _mm256_mul_ps(fscal,dy00);
 294             tz               = _mm256_mul_ps(fscal,dz00);
 295
 296             /* Update vectorial force */
 297             fix0             = _mm256_add_ps(fix0,tx);
 298             fiy0             = _mm256_add_ps(fiy0,ty);
 299             fiz0             = _mm256_add_ps(fiz0,tz);
 300
 301             fjx0             = _mm256_add_ps(fjx0,tx);
 302             fjy0             = _mm256_add_ps(fjy0,ty);
 303             fjz0             = _mm256_add_ps(fjz0,tz);
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             /* Compute parameters for interactions between i and j atoms */
 310             qq10             = _mm256_mul_ps(iq1,jq0);
 311
 312             /* COULOMB ELECTROSTATICS */
 313             velec            = _mm256_mul_ps(qq10,rinv10);
 314             felec            = _mm256_mul_ps(velec,rinvsq10);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velecsum         = _mm256_add_ps(velecsum,velec);
 318
 319             fscal            = felec;
 320
 321             /* Calculate temporary vectorial force */
 322             tx               = _mm256_mul_ps(fscal,dx10);
 323             ty               = _mm256_mul_ps(fscal,dy10);
 324             tz               = _mm256_mul_ps(fscal,dz10);
 325
 326             /* Update vectorial force */
 327             fix1             = _mm256_add_ps(fix1,tx);
 328             fiy1             = _mm256_add_ps(fiy1,ty);
 329             fiz1             = _mm256_add_ps(fiz1,tz);
 330
 331             fjx0             = _mm256_add_ps(fjx0,tx);
 332             fjy0             = _mm256_add_ps(fjy0,ty);
 333             fjz0             = _mm256_add_ps(fjz0,tz);
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq20             = _mm256_mul_ps(iq2,jq0);
 341
 342             /* COULOMB ELECTROSTATICS */
 343             velec            = _mm256_mul_ps(qq20,rinv20);
 344             felec            = _mm256_mul_ps(velec,rinvsq20);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm256_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm256_mul_ps(fscal,dx20);
 353             ty               = _mm256_mul_ps(fscal,dy20);
 354             tz               = _mm256_mul_ps(fscal,dz20);
 355
 356             /* Update vectorial force */
 357             fix2             = _mm256_add_ps(fix2,tx);
 358             fiy2             = _mm256_add_ps(fiy2,ty);
 359             fiz2             = _mm256_add_ps(fiz2,tz);
 360
 361             fjx0             = _mm256_add_ps(fjx0,tx);
 362             fjy0             = _mm256_add_ps(fjy0,ty);
 363             fjz0             = _mm256_add_ps(fjz0,tz);
 364
 365             fjptrA             = f+j_coord_offsetA;
 366             fjptrB             = f+j_coord_offsetB;
 367             fjptrC             = f+j_coord_offsetC;
 368             fjptrD             = f+j_coord_offsetD;
 369             fjptrE             = f+j_coord_offsetE;
 370             fjptrF             = f+j_coord_offsetF;
 371             fjptrG             = f+j_coord_offsetG;
 372             fjptrH             = f+j_coord_offsetH;
 373
 374             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 375
 376             /* Inner loop uses 96 flops */
 377         }
 378
 379         if(jidx<j_index_end)
 380         {
 381
 382             /* Get j neighbor index, and coordinate index */
 383             jnrlistA         = jjnr[jidx];
 384             jnrlistB         = jjnr[jidx+1];
 385             jnrlistC         = jjnr[jidx+2];
 386             jnrlistD         = jjnr[jidx+3];
 387             jnrlistE         = jjnr[jidx+4];
 388             jnrlistF         = jjnr[jidx+5];
 389             jnrlistG         = jjnr[jidx+6];
 390             jnrlistH         = jjnr[jidx+7];
 391             /* Sign of each element will be negative for non-real atoms.
 392              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 393              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 394              */
 395             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 396                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 397
 398             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 399             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 400             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 401             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 402             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 403             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 404             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 405             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 406             j_coord_offsetA  = DIM*jnrA;
 407             j_coord_offsetB  = DIM*jnrB;
 408             j_coord_offsetC  = DIM*jnrC;
 409             j_coord_offsetD  = DIM*jnrD;
 410             j_coord_offsetE  = DIM*jnrE;
 411             j_coord_offsetF  = DIM*jnrF;
 412             j_coord_offsetG  = DIM*jnrG;
 413             j_coord_offsetH  = DIM*jnrH;
 414
 415             /* load j atom coordinates */
 416             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 417                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 418                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 419                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 420                                                  &jx0,&jy0,&jz0);
 421
 422             /* Calculate displacement vector */
 423             dx00             = _mm256_sub_ps(ix0,jx0);
 424             dy00             = _mm256_sub_ps(iy0,jy0);
 425             dz00             = _mm256_sub_ps(iz0,jz0);
 426             dx10             = _mm256_sub_ps(ix1,jx0);
 427             dy10             = _mm256_sub_ps(iy1,jy0);
 428             dz10             = _mm256_sub_ps(iz1,jz0);
 429             dx20             = _mm256_sub_ps(ix2,jx0);
 430             dy20             = _mm256_sub_ps(iy2,jy0);
 431             dz20             = _mm256_sub_ps(iz2,jz0);
 432
 433             /* Calculate squared distance and things based on it */
 434             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 435             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 436             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 437
 438             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 439             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 440             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 441
 442             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 443             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 444             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 445
 446             /* Load parameters for j particles */
 447             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 448                                                                  charge+jnrC+0,charge+jnrD+0,
 449                                                                  charge+jnrE+0,charge+jnrF+0,
 450                                                                  charge+jnrG+0,charge+jnrH+0);
 451             vdwjidx0A        = 2*vdwtype[jnrA+0];
 452             vdwjidx0B        = 2*vdwtype[jnrB+0];
 453             vdwjidx0C        = 2*vdwtype[jnrC+0];
 454             vdwjidx0D        = 2*vdwtype[jnrD+0];
 455             vdwjidx0E        = 2*vdwtype[jnrE+0];
 456             vdwjidx0F        = 2*vdwtype[jnrF+0];
 457             vdwjidx0G        = 2*vdwtype[jnrG+0];
 458             vdwjidx0H        = 2*vdwtype[jnrH+0];
 459
 460             fjx0             = _mm256_setzero_ps();
 461             fjy0             = _mm256_setzero_ps();
 462             fjz0             = _mm256_setzero_ps();
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             /* Compute parameters for interactions between i and j atoms */
 469             qq00             = _mm256_mul_ps(iq0,jq0);
 470             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 471                                             vdwioffsetptr0+vdwjidx0B,
 472                                             vdwioffsetptr0+vdwjidx0C,
 473                                             vdwioffsetptr0+vdwjidx0D,
 474                                             vdwioffsetptr0+vdwjidx0E,
 475                                             vdwioffsetptr0+vdwjidx0F,
 476                                             vdwioffsetptr0+vdwjidx0G,
 477                                             vdwioffsetptr0+vdwjidx0H,
 478                                             &c6_00,&c12_00);
 479
 480             /* COULOMB ELECTROSTATICS */
 481             velec            = _mm256_mul_ps(qq00,rinv00);
 482             felec            = _mm256_mul_ps(velec,rinvsq00);
 483
 484             /* LENNARD-JONES DISPERSION/REPULSION */
 485
 486             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 487             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 488             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 489             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 490             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             velec            = _mm256_andnot_ps(dummy_mask,velec);
 494             velecsum         = _mm256_add_ps(velecsum,velec);
 495             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 496             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 497
 498             fscal            = _mm256_add_ps(felec,fvdw);
 499
 500             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 501
 502             /* Calculate temporary vectorial force */
 503             tx               = _mm256_mul_ps(fscal,dx00);
 504             ty               = _mm256_mul_ps(fscal,dy00);
 505             tz               = _mm256_mul_ps(fscal,dz00);
 506
 507             /* Update vectorial force */
 508             fix0             = _mm256_add_ps(fix0,tx);
 509             fiy0             = _mm256_add_ps(fiy0,ty);
 510             fiz0             = _mm256_add_ps(fiz0,tz);
 511
 512             fjx0             = _mm256_add_ps(fjx0,tx);
 513             fjy0             = _mm256_add_ps(fjy0,ty);
 514             fjz0             = _mm256_add_ps(fjz0,tz);
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             qq10             = _mm256_mul_ps(iq1,jq0);
 522
 523             /* COULOMB ELECTROSTATICS */
 524             velec            = _mm256_mul_ps(qq10,rinv10);
 525             felec            = _mm256_mul_ps(velec,rinvsq10);
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             velec            = _mm256_andnot_ps(dummy_mask,velec);
 529             velecsum         = _mm256_add_ps(velecsum,velec);
 530
 531             fscal            = felec;
 532
 533             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 534
 535             /* Calculate temporary vectorial force */
 536             tx               = _mm256_mul_ps(fscal,dx10);
 537             ty               = _mm256_mul_ps(fscal,dy10);
 538             tz               = _mm256_mul_ps(fscal,dz10);
 539
 540             /* Update vectorial force */
 541             fix1             = _mm256_add_ps(fix1,tx);
 542             fiy1             = _mm256_add_ps(fiy1,ty);
 543             fiz1             = _mm256_add_ps(fiz1,tz);
 544
 545             fjx0             = _mm256_add_ps(fjx0,tx);
 546             fjy0             = _mm256_add_ps(fjy0,ty);
 547             fjz0             = _mm256_add_ps(fjz0,tz);
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq20             = _mm256_mul_ps(iq2,jq0);
 555
 556             /* COULOMB ELECTROSTATICS */
 557             velec            = _mm256_mul_ps(qq20,rinv20);
 558             felec            = _mm256_mul_ps(velec,rinvsq20);
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velec            = _mm256_andnot_ps(dummy_mask,velec);
 562             velecsum         = _mm256_add_ps(velecsum,velec);
 563
 564             fscal            = felec;
 565
 566             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm256_mul_ps(fscal,dx20);
 570             ty               = _mm256_mul_ps(fscal,dy20);
 571             tz               = _mm256_mul_ps(fscal,dz20);
 572
 573             /* Update vectorial force */
 574             fix2             = _mm256_add_ps(fix2,tx);
 575             fiy2             = _mm256_add_ps(fiy2,ty);
 576             fiz2             = _mm256_add_ps(fiz2,tz);
 577
 578             fjx0             = _mm256_add_ps(fjx0,tx);
 579             fjy0             = _mm256_add_ps(fjy0,ty);
 580             fjz0             = _mm256_add_ps(fjz0,tz);
 581
 582             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 583             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 584             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 585             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 586             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 587             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 588             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 589             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 590
 591             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 592
 593             /* Inner loop uses 96 flops */
 594         }
 595
 596         /* End of innermost loop */
 597
 598         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 599                                                  f+i_coord_offset,fshift+i_shift_offset);
 600
 601         ggid                        = gid[iidx];
 602         /* Update potential energies */
 603         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 604         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 605
 606         /* Increment number of inner iterations */
 607         inneriter                  += j_index_end - j_index_start;
 608
 609         /* Outer loop uses 20 flops */
 610     }
 611
 612     /* Increment number of outer iterations */
 613     outeriter        += nri;
 614
 615     /* Update outer/inner flops */
 616
 617     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
 618 }
 619 /*
 620  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_single
 621  * Electrostatics interaction: Coulomb
 622  * VdW interaction:            LennardJones
 623  * Geometry:                   Water3-Particle
 624  * Calculate force/pot:        Force
 625  */
 626 void
 627 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_single
 628                     (t_nblist                    * gmx_restrict       nlist,
 629                      rvec                        * gmx_restrict          xx,
 630                      rvec                        * gmx_restrict          ff,
 631                      t_forcerec                  * gmx_restrict          fr,
 632                      t_mdatoms                   * gmx_restrict     mdatoms,
 633                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 634                      t_nrnb                      * gmx_restrict        nrnb)
 635 {
 636     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 637      * just 0 for non-waters.
 638      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 639      * jnr indices corresponding to data put in the four positions in the SIMD register.
 640      */
 641     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 642     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 643     int              jnrA,jnrB,jnrC,jnrD;
 644     int              jnrE,jnrF,jnrG,jnrH;
 645     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 646     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 647     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 648     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 649     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 650     real             rcutoff_scalar;
 651     real             *shiftvec,*fshift,*x,*f;
 652     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 653     real             scratch[4*DIM];
 654     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 655     real *           vdwioffsetptr0;
 656     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 657     real *           vdwioffsetptr1;
 658     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 659     real *           vdwioffsetptr2;
 660     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 661     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 662     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 663     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 664     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 665     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 666     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 667     real             *charge;
 668     int              nvdwtype;
 669     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 670     int              *vdwtype;
 671     real             *vdwparam;
 672     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 673     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 674     __m256           dummy_mask,cutoff_mask;
 675     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 676     __m256           one     = _mm256_set1_ps(1.0);
 677     __m256           two     = _mm256_set1_ps(2.0);
 678     x                = xx[0];
 679     f                = ff[0];
 680
 681     nri              = nlist->nri;
 682     iinr             = nlist->iinr;
 683     jindex           = nlist->jindex;
 684     jjnr             = nlist->jjnr;
 685     shiftidx         = nlist->shift;
 686     gid              = nlist->gid;
 687     shiftvec         = fr->shift_vec[0];
 688     fshift           = fr->fshift[0];
 689     facel            = _mm256_set1_ps(fr->epsfac);
 690     charge           = mdatoms->chargeA;
 691     nvdwtype         = fr->ntype;
 692     vdwparam         = fr->nbfp;
 693     vdwtype          = mdatoms->typeA;
 694
 695     /* Setup water-specific parameters */
 696     inr              = nlist->iinr[0];
 697     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 698     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 699     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 700     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 701
 702     /* Avoid stupid compiler warnings */
 703     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 704     j_coord_offsetA = 0;
 705     j_coord_offsetB = 0;
 706     j_coord_offsetC = 0;
 707     j_coord_offsetD = 0;
 708     j_coord_offsetE = 0;
 709     j_coord_offsetF = 0;
 710     j_coord_offsetG = 0;
 711     j_coord_offsetH = 0;
 712
 713     outeriter        = 0;
 714     inneriter        = 0;
 715
 716     for(iidx=0;iidx<4*DIM;iidx++)
 717     {
 718         scratch[iidx] = 0.0;
 719     }
 720
 721     /* Start outer loop over neighborlists */
 722     for(iidx=0; iidx<nri; iidx++)
 723     {
 724         /* Load shift vector for this list */
 725         i_shift_offset   = DIM*shiftidx[iidx];
 726
 727         /* Load limits for loop over neighbors */
 728         j_index_start    = jindex[iidx];
 729         j_index_end      = jindex[iidx+1];
 730
 731         /* Get outer coordinate index */
 732         inr              = iinr[iidx];
 733         i_coord_offset   = DIM*inr;
 734
 735         /* Load i particle coords and add shift vector */
 736         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 737                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 738
 739         fix0             = _mm256_setzero_ps();
 740         fiy0             = _mm256_setzero_ps();
 741         fiz0             = _mm256_setzero_ps();
 742         fix1             = _mm256_setzero_ps();
 743         fiy1             = _mm256_setzero_ps();
 744         fiz1             = _mm256_setzero_ps();
 745         fix2             = _mm256_setzero_ps();
 746         fiy2             = _mm256_setzero_ps();
 747         fiz2             = _mm256_setzero_ps();
 748
 749         /* Start inner kernel loop */
 750         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 751         {
 752
 753             /* Get j neighbor index, and coordinate index */
 754             jnrA             = jjnr[jidx];
 755             jnrB             = jjnr[jidx+1];
 756             jnrC             = jjnr[jidx+2];
 757             jnrD             = jjnr[jidx+3];
 758             jnrE             = jjnr[jidx+4];
 759             jnrF             = jjnr[jidx+5];
 760             jnrG             = jjnr[jidx+6];
 761             jnrH             = jjnr[jidx+7];
 762             j_coord_offsetA  = DIM*jnrA;
 763             j_coord_offsetB  = DIM*jnrB;
 764             j_coord_offsetC  = DIM*jnrC;
 765             j_coord_offsetD  = DIM*jnrD;
 766             j_coord_offsetE  = DIM*jnrE;
 767             j_coord_offsetF  = DIM*jnrF;
 768             j_coord_offsetG  = DIM*jnrG;
 769             j_coord_offsetH  = DIM*jnrH;
 770
 771             /* load j atom coordinates */
 772             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 773                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 774                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 775                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 776                                                  &jx0,&jy0,&jz0);
 777
 778             /* Calculate displacement vector */
 779             dx00             = _mm256_sub_ps(ix0,jx0);
 780             dy00             = _mm256_sub_ps(iy0,jy0);
 781             dz00             = _mm256_sub_ps(iz0,jz0);
 782             dx10             = _mm256_sub_ps(ix1,jx0);
 783             dy10             = _mm256_sub_ps(iy1,jy0);
 784             dz10             = _mm256_sub_ps(iz1,jz0);
 785             dx20             = _mm256_sub_ps(ix2,jx0);
 786             dy20             = _mm256_sub_ps(iy2,jy0);
 787             dz20             = _mm256_sub_ps(iz2,jz0);
 788
 789             /* Calculate squared distance and things based on it */
 790             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 791             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 792             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 793
 794             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 795             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 796             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 797
 798             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 799             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 800             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 801
 802             /* Load parameters for j particles */
 803             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 804                                                                  charge+jnrC+0,charge+jnrD+0,
 805                                                                  charge+jnrE+0,charge+jnrF+0,
 806                                                                  charge+jnrG+0,charge+jnrH+0);
 807             vdwjidx0A        = 2*vdwtype[jnrA+0];
 808             vdwjidx0B        = 2*vdwtype[jnrB+0];
 809             vdwjidx0C        = 2*vdwtype[jnrC+0];
 810             vdwjidx0D        = 2*vdwtype[jnrD+0];
 811             vdwjidx0E        = 2*vdwtype[jnrE+0];
 812             vdwjidx0F        = 2*vdwtype[jnrF+0];
 813             vdwjidx0G        = 2*vdwtype[jnrG+0];
 814             vdwjidx0H        = 2*vdwtype[jnrH+0];
 815
 816             fjx0             = _mm256_setzero_ps();
 817             fjy0             = _mm256_setzero_ps();
 818             fjz0             = _mm256_setzero_ps();
 819
 820             /**************************
 821              * CALCULATE INTERACTIONS *
 822              **************************/
 823
 824             /* Compute parameters for interactions between i and j atoms */
 825             qq00             = _mm256_mul_ps(iq0,jq0);
 826             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 827                                             vdwioffsetptr0+vdwjidx0B,
 828                                             vdwioffsetptr0+vdwjidx0C,
 829                                             vdwioffsetptr0+vdwjidx0D,
 830                                             vdwioffsetptr0+vdwjidx0E,
 831                                             vdwioffsetptr0+vdwjidx0F,
 832                                             vdwioffsetptr0+vdwjidx0G,
 833                                             vdwioffsetptr0+vdwjidx0H,
 834                                             &c6_00,&c12_00);
 835
 836             /* COULOMB ELECTROSTATICS */
 837             velec            = _mm256_mul_ps(qq00,rinv00);
 838             felec            = _mm256_mul_ps(velec,rinvsq00);
 839
 840             /* LENNARD-JONES DISPERSION/REPULSION */
 841
 842             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 843             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 844
 845             fscal            = _mm256_add_ps(felec,fvdw);
 846
 847             /* Calculate temporary vectorial force */
 848             tx               = _mm256_mul_ps(fscal,dx00);
 849             ty               = _mm256_mul_ps(fscal,dy00);
 850             tz               = _mm256_mul_ps(fscal,dz00);
 851
 852             /* Update vectorial force */
 853             fix0             = _mm256_add_ps(fix0,tx);
 854             fiy0             = _mm256_add_ps(fiy0,ty);
 855             fiz0             = _mm256_add_ps(fiz0,tz);
 856
 857             fjx0             = _mm256_add_ps(fjx0,tx);
 858             fjy0             = _mm256_add_ps(fjy0,ty);
 859             fjz0             = _mm256_add_ps(fjz0,tz);
 860
 861             /**************************
 862              * CALCULATE INTERACTIONS *
 863              **************************/
 864
 865             /* Compute parameters for interactions between i and j atoms */
 866             qq10             = _mm256_mul_ps(iq1,jq0);
 867
 868             /* COULOMB ELECTROSTATICS */
 869             velec            = _mm256_mul_ps(qq10,rinv10);
 870             felec            = _mm256_mul_ps(velec,rinvsq10);
 871
 872             fscal            = felec;
 873
 874             /* Calculate temporary vectorial force */
 875             tx               = _mm256_mul_ps(fscal,dx10);
 876             ty               = _mm256_mul_ps(fscal,dy10);
 877             tz               = _mm256_mul_ps(fscal,dz10);
 878
 879             /* Update vectorial force */
 880             fix1             = _mm256_add_ps(fix1,tx);
 881             fiy1             = _mm256_add_ps(fiy1,ty);
 882             fiz1             = _mm256_add_ps(fiz1,tz);
 883
 884             fjx0             = _mm256_add_ps(fjx0,tx);
 885             fjy0             = _mm256_add_ps(fjy0,ty);
 886             fjz0             = _mm256_add_ps(fjz0,tz);
 887
 888             /**************************
 889              * CALCULATE INTERACTIONS *
 890              **************************/
 891
 892             /* Compute parameters for interactions between i and j atoms */
 893             qq20             = _mm256_mul_ps(iq2,jq0);
 894
 895             /* COULOMB ELECTROSTATICS */
 896             velec            = _mm256_mul_ps(qq20,rinv20);
 897             felec            = _mm256_mul_ps(velec,rinvsq20);
 898
 899             fscal            = felec;
 900
 901             /* Calculate temporary vectorial force */
 902             tx               = _mm256_mul_ps(fscal,dx20);
 903             ty               = _mm256_mul_ps(fscal,dy20);
 904             tz               = _mm256_mul_ps(fscal,dz20);
 905
 906             /* Update vectorial force */
 907             fix2             = _mm256_add_ps(fix2,tx);
 908             fiy2             = _mm256_add_ps(fiy2,ty);
 909             fiz2             = _mm256_add_ps(fiz2,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             fjptrA             = f+j_coord_offsetA;
 916             fjptrB             = f+j_coord_offsetB;
 917             fjptrC             = f+j_coord_offsetC;
 918             fjptrD             = f+j_coord_offsetD;
 919             fjptrE             = f+j_coord_offsetE;
 920             fjptrF             = f+j_coord_offsetF;
 921             fjptrG             = f+j_coord_offsetG;
 922             fjptrH             = f+j_coord_offsetH;
 923
 924             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 925
 926             /* Inner loop uses 88 flops */
 927         }
 928
 929         if(jidx<j_index_end)
 930         {
 931
 932             /* Get j neighbor index, and coordinate index */
 933             jnrlistA         = jjnr[jidx];
 934             jnrlistB         = jjnr[jidx+1];
 935             jnrlistC         = jjnr[jidx+2];
 936             jnrlistD         = jjnr[jidx+3];
 937             jnrlistE         = jjnr[jidx+4];
 938             jnrlistF         = jjnr[jidx+5];
 939             jnrlistG         = jjnr[jidx+6];
 940             jnrlistH         = jjnr[jidx+7];
 941             /* Sign of each element will be negative for non-real atoms.
 942              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 943              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 944              */
 945             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 946                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 947
 948             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 949             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 950             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 951             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 952             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 953             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 954             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 955             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 956             j_coord_offsetA  = DIM*jnrA;
 957             j_coord_offsetB  = DIM*jnrB;
 958             j_coord_offsetC  = DIM*jnrC;
 959             j_coord_offsetD  = DIM*jnrD;
 960             j_coord_offsetE  = DIM*jnrE;
 961             j_coord_offsetF  = DIM*jnrF;
 962             j_coord_offsetG  = DIM*jnrG;
 963             j_coord_offsetH  = DIM*jnrH;
 964
 965             /* load j atom coordinates */
 966             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 967                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 968                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 969                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 970                                                  &jx0,&jy0,&jz0);
 971
 972             /* Calculate displacement vector */
 973             dx00             = _mm256_sub_ps(ix0,jx0);
 974             dy00             = _mm256_sub_ps(iy0,jy0);
 975             dz00             = _mm256_sub_ps(iz0,jz0);
 976             dx10             = _mm256_sub_ps(ix1,jx0);
 977             dy10             = _mm256_sub_ps(iy1,jy0);
 978             dz10             = _mm256_sub_ps(iz1,jz0);
 979             dx20             = _mm256_sub_ps(ix2,jx0);
 980             dy20             = _mm256_sub_ps(iy2,jy0);
 981             dz20             = _mm256_sub_ps(iz2,jz0);
 982
 983             /* Calculate squared distance and things based on it */
 984             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 985             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 986             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 987
 988             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 989             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 990             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 991
 992             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 993             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 994             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 995
 996             /* Load parameters for j particles */
 997             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 998                                                                  charge+jnrC+0,charge+jnrD+0,
 999                                                                  charge+jnrE+0,charge+jnrF+0,
1000                                                                  charge+jnrG+0,charge+jnrH+0);
1001             vdwjidx0A        = 2*vdwtype[jnrA+0];
1002             vdwjidx0B        = 2*vdwtype[jnrB+0];
1003             vdwjidx0C        = 2*vdwtype[jnrC+0];
1004             vdwjidx0D        = 2*vdwtype[jnrD+0];
1005             vdwjidx0E        = 2*vdwtype[jnrE+0];
1006             vdwjidx0F        = 2*vdwtype[jnrF+0];
1007             vdwjidx0G        = 2*vdwtype[jnrG+0];
1008             vdwjidx0H        = 2*vdwtype[jnrH+0];
1009
1010             fjx0             = _mm256_setzero_ps();
1011             fjy0             = _mm256_setzero_ps();
1012             fjz0             = _mm256_setzero_ps();
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             /* Compute parameters for interactions between i and j atoms */
1019             qq00             = _mm256_mul_ps(iq0,jq0);
1020             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1021                                             vdwioffsetptr0+vdwjidx0B,
1022                                             vdwioffsetptr0+vdwjidx0C,
1023                                             vdwioffsetptr0+vdwjidx0D,
1024                                             vdwioffsetptr0+vdwjidx0E,
1025                                             vdwioffsetptr0+vdwjidx0F,
1026                                             vdwioffsetptr0+vdwjidx0G,
1027                                             vdwioffsetptr0+vdwjidx0H,
1028                                             &c6_00,&c12_00);
1029
1030             /* COULOMB ELECTROSTATICS */
1031             velec            = _mm256_mul_ps(qq00,rinv00);
1032             felec            = _mm256_mul_ps(velec,rinvsq00);
1033
1034             /* LENNARD-JONES DISPERSION/REPULSION */
1035
1036             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1037             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1038
1039             fscal            = _mm256_add_ps(felec,fvdw);
1040
1041             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1042
1043             /* Calculate temporary vectorial force */
1044             tx               = _mm256_mul_ps(fscal,dx00);
1045             ty               = _mm256_mul_ps(fscal,dy00);
1046             tz               = _mm256_mul_ps(fscal,dz00);
1047
1048             /* Update vectorial force */
1049             fix0             = _mm256_add_ps(fix0,tx);
1050             fiy0             = _mm256_add_ps(fiy0,ty);
1051             fiz0             = _mm256_add_ps(fiz0,tz);
1052
1053             fjx0             = _mm256_add_ps(fjx0,tx);
1054             fjy0             = _mm256_add_ps(fjy0,ty);
1055             fjz0             = _mm256_add_ps(fjz0,tz);
1056
1057             /**************************
1058              * CALCULATE INTERACTIONS *
1059              **************************/
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq10             = _mm256_mul_ps(iq1,jq0);
1063
1064             /* COULOMB ELECTROSTATICS */
1065             velec            = _mm256_mul_ps(qq10,rinv10);
1066             felec            = _mm256_mul_ps(velec,rinvsq10);
1067
1068             fscal            = felec;
1069
1070             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1071
1072             /* Calculate temporary vectorial force */
1073             tx               = _mm256_mul_ps(fscal,dx10);
1074             ty               = _mm256_mul_ps(fscal,dy10);
1075             tz               = _mm256_mul_ps(fscal,dz10);
1076
1077             /* Update vectorial force */
1078             fix1             = _mm256_add_ps(fix1,tx);
1079             fiy1             = _mm256_add_ps(fiy1,ty);
1080             fiz1             = _mm256_add_ps(fiz1,tz);
1081
1082             fjx0             = _mm256_add_ps(fjx0,tx);
1083             fjy0             = _mm256_add_ps(fjy0,ty);
1084             fjz0             = _mm256_add_ps(fjz0,tz);
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             /* Compute parameters for interactions between i and j atoms */
1091             qq20             = _mm256_mul_ps(iq2,jq0);
1092
1093             /* COULOMB ELECTROSTATICS */
1094             velec            = _mm256_mul_ps(qq20,rinv20);
1095             felec            = _mm256_mul_ps(velec,rinvsq20);
1096
1097             fscal            = felec;
1098
1099             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1100
1101             /* Calculate temporary vectorial force */
1102             tx               = _mm256_mul_ps(fscal,dx20);
1103             ty               = _mm256_mul_ps(fscal,dy20);
1104             tz               = _mm256_mul_ps(fscal,dz20);
1105
1106             /* Update vectorial force */
1107             fix2             = _mm256_add_ps(fix2,tx);
1108             fiy2             = _mm256_add_ps(fiy2,ty);
1109             fiz2             = _mm256_add_ps(fiz2,tz);
1110
1111             fjx0             = _mm256_add_ps(fjx0,tx);
1112             fjy0             = _mm256_add_ps(fjy0,ty);
1113             fjz0             = _mm256_add_ps(fjz0,tz);
1114
1115             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1116             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1117             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1118             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1119             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1120             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1121             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1122             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1123
1124             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1125
1126             /* Inner loop uses 88 flops */
1127         }
1128
1129         /* End of innermost loop */
1130
1131         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1132                                                  f+i_coord_offset,fshift+i_shift_offset);
1133
1134         /* Increment number of inner iterations */
1135         inneriter                  += j_index_end - j_index_start;
1136
1137         /* Outer loop uses 18 flops */
1138     }
1139
1140     /* Increment number of outer iterations */
1141     outeriter        += nri;
1142
1143     /* Update outer/inner flops */
1144
1145     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);
1146 }