src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwLJ_GeomW4P1_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_GeomW4P1_VF_avx_256_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_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     real *           vdwioffsetptr3;
  95     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  96     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  97     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 100     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 102     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 103     real             *charge;
 104     int              nvdwtype;
 105     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 106     int              *vdwtype;
 107     real             *vdwparam;
 108     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 109     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 110     __m256           dummy_mask,cutoff_mask;
 111     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 112     __m256           one     = _mm256_set1_ps(1.0);
 113     __m256           two     = _mm256_set1_ps(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_ps(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 134     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 135     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 136     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144     j_coord_offsetE = 0;
 145     j_coord_offsetF = 0;
 146     j_coord_offsetG = 0;
 147     j_coord_offsetH = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 174
 175         fix0             = _mm256_setzero_ps();
 176         fiy0             = _mm256_setzero_ps();
 177         fiz0             = _mm256_setzero_ps();
 178         fix1             = _mm256_setzero_ps();
 179         fiy1             = _mm256_setzero_ps();
 180         fiz1             = _mm256_setzero_ps();
 181         fix2             = _mm256_setzero_ps();
 182         fiy2             = _mm256_setzero_ps();
 183         fiz2             = _mm256_setzero_ps();
 184         fix3             = _mm256_setzero_ps();
 185         fiy3             = _mm256_setzero_ps();
 186         fiz3             = _mm256_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_ps();
 190         vvdwsum          = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225             dx10             = _mm256_sub_ps(ix1,jx0);
 226             dy10             = _mm256_sub_ps(iy1,jy0);
 227             dz10             = _mm256_sub_ps(iz1,jz0);
 228             dx20             = _mm256_sub_ps(ix2,jx0);
 229             dy20             = _mm256_sub_ps(iy2,jy0);
 230             dz20             = _mm256_sub_ps(iz2,jz0);
 231             dx30             = _mm256_sub_ps(ix3,jx0);
 232             dy30             = _mm256_sub_ps(iy3,jy0);
 233             dz30             = _mm256_sub_ps(iz3,jz0);
 234
 235             /* Calculate squared distance and things based on it */
 236             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 237             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 238             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 239             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 240
 241             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 242             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 243             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 244
 245             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 246             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 247             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 248             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 249
 250             /* Load parameters for j particles */
 251             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 252                                                                  charge+jnrC+0,charge+jnrD+0,
 253                                                                  charge+jnrE+0,charge+jnrF+0,
 254                                                                  charge+jnrG+0,charge+jnrH+0);
 255             vdwjidx0A        = 2*vdwtype[jnrA+0];
 256             vdwjidx0B        = 2*vdwtype[jnrB+0];
 257             vdwjidx0C        = 2*vdwtype[jnrC+0];
 258             vdwjidx0D        = 2*vdwtype[jnrD+0];
 259             vdwjidx0E        = 2*vdwtype[jnrE+0];
 260             vdwjidx0F        = 2*vdwtype[jnrF+0];
 261             vdwjidx0G        = 2*vdwtype[jnrG+0];
 262             vdwjidx0H        = 2*vdwtype[jnrH+0];
 263
 264             fjx0             = _mm256_setzero_ps();
 265             fjy0             = _mm256_setzero_ps();
 266             fjz0             = _mm256_setzero_ps();
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 274                                             vdwioffsetptr0+vdwjidx0B,
 275                                             vdwioffsetptr0+vdwjidx0C,
 276                                             vdwioffsetptr0+vdwjidx0D,
 277                                             vdwioffsetptr0+vdwjidx0E,
 278                                             vdwioffsetptr0+vdwjidx0F,
 279                                             vdwioffsetptr0+vdwjidx0G,
 280                                             vdwioffsetptr0+vdwjidx0H,
 281                                             &c6_00,&c12_00);
 282
 283             /* LENNARD-JONES DISPERSION/REPULSION */
 284
 285             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 286             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 287             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 288             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 289             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 290
 291             /* Update potential sum for this i atom from the interaction with this j atom. */
 292             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 293
 294             fscal            = fvdw;
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm256_mul_ps(fscal,dx00);
 298             ty               = _mm256_mul_ps(fscal,dy00);
 299             tz               = _mm256_mul_ps(fscal,dz00);
 300
 301             /* Update vectorial force */
 302             fix0             = _mm256_add_ps(fix0,tx);
 303             fiy0             = _mm256_add_ps(fiy0,ty);
 304             fiz0             = _mm256_add_ps(fiz0,tz);
 305
 306             fjx0             = _mm256_add_ps(fjx0,tx);
 307             fjy0             = _mm256_add_ps(fjy0,ty);
 308             fjz0             = _mm256_add_ps(fjz0,tz);
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq10             = _mm256_mul_ps(iq1,jq0);
 316
 317             /* COULOMB ELECTROSTATICS */
 318             velec            = _mm256_mul_ps(qq10,rinv10);
 319             felec            = _mm256_mul_ps(velec,rinvsq10);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velecsum         = _mm256_add_ps(velecsum,velec);
 323
 324             fscal            = felec;
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm256_mul_ps(fscal,dx10);
 328             ty               = _mm256_mul_ps(fscal,dy10);
 329             tz               = _mm256_mul_ps(fscal,dz10);
 330
 331             /* Update vectorial force */
 332             fix1             = _mm256_add_ps(fix1,tx);
 333             fiy1             = _mm256_add_ps(fiy1,ty);
 334             fiz1             = _mm256_add_ps(fiz1,tz);
 335
 336             fjx0             = _mm256_add_ps(fjx0,tx);
 337             fjy0             = _mm256_add_ps(fjy0,ty);
 338             fjz0             = _mm256_add_ps(fjz0,tz);
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq20             = _mm256_mul_ps(iq2,jq0);
 346
 347             /* COULOMB ELECTROSTATICS */
 348             velec            = _mm256_mul_ps(qq20,rinv20);
 349             felec            = _mm256_mul_ps(velec,rinvsq20);
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velecsum         = _mm256_add_ps(velecsum,velec);
 353
 354             fscal            = felec;
 355
 356             /* Calculate temporary vectorial force */
 357             tx               = _mm256_mul_ps(fscal,dx20);
 358             ty               = _mm256_mul_ps(fscal,dy20);
 359             tz               = _mm256_mul_ps(fscal,dz20);
 360
 361             /* Update vectorial force */
 362             fix2             = _mm256_add_ps(fix2,tx);
 363             fiy2             = _mm256_add_ps(fiy2,ty);
 364             fiz2             = _mm256_add_ps(fiz2,tz);
 365
 366             fjx0             = _mm256_add_ps(fjx0,tx);
 367             fjy0             = _mm256_add_ps(fjy0,ty);
 368             fjz0             = _mm256_add_ps(fjz0,tz);
 369
 370             /**************************
 371              * CALCULATE INTERACTIONS *
 372              **************************/
 373
 374             /* Compute parameters for interactions between i and j atoms */
 375             qq30             = _mm256_mul_ps(iq3,jq0);
 376
 377             /* COULOMB ELECTROSTATICS */
 378             velec            = _mm256_mul_ps(qq30,rinv30);
 379             felec            = _mm256_mul_ps(velec,rinvsq30);
 380
 381             /* Update potential sum for this i atom from the interaction with this j atom. */
 382             velecsum         = _mm256_add_ps(velecsum,velec);
 383
 384             fscal            = felec;
 385
 386             /* Calculate temporary vectorial force */
 387             tx               = _mm256_mul_ps(fscal,dx30);
 388             ty               = _mm256_mul_ps(fscal,dy30);
 389             tz               = _mm256_mul_ps(fscal,dz30);
 390
 391             /* Update vectorial force */
 392             fix3             = _mm256_add_ps(fix3,tx);
 393             fiy3             = _mm256_add_ps(fiy3,ty);
 394             fiz3             = _mm256_add_ps(fiz3,tz);
 395
 396             fjx0             = _mm256_add_ps(fjx0,tx);
 397             fjy0             = _mm256_add_ps(fjy0,ty);
 398             fjz0             = _mm256_add_ps(fjz0,tz);
 399
 400             fjptrA             = f+j_coord_offsetA;
 401             fjptrB             = f+j_coord_offsetB;
 402             fjptrC             = f+j_coord_offsetC;
 403             fjptrD             = f+j_coord_offsetD;
 404             fjptrE             = f+j_coord_offsetE;
 405             fjptrF             = f+j_coord_offsetF;
 406             fjptrG             = f+j_coord_offsetG;
 407             fjptrH             = f+j_coord_offsetH;
 408
 409             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 410
 411             /* Inner loop uses 116 flops */
 412         }
 413
 414         if(jidx<j_index_end)
 415         {
 416
 417             /* Get j neighbor index, and coordinate index */
 418             jnrlistA         = jjnr[jidx];
 419             jnrlistB         = jjnr[jidx+1];
 420             jnrlistC         = jjnr[jidx+2];
 421             jnrlistD         = jjnr[jidx+3];
 422             jnrlistE         = jjnr[jidx+4];
 423             jnrlistF         = jjnr[jidx+5];
 424             jnrlistG         = jjnr[jidx+6];
 425             jnrlistH         = jjnr[jidx+7];
 426             /* Sign of each element will be negative for non-real atoms.
 427              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 428              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 429              */
 430             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 431                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 432
 433             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 434             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 435             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 436             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 437             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 438             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 439             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 440             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 441             j_coord_offsetA  = DIM*jnrA;
 442             j_coord_offsetB  = DIM*jnrB;
 443             j_coord_offsetC  = DIM*jnrC;
 444             j_coord_offsetD  = DIM*jnrD;
 445             j_coord_offsetE  = DIM*jnrE;
 446             j_coord_offsetF  = DIM*jnrF;
 447             j_coord_offsetG  = DIM*jnrG;
 448             j_coord_offsetH  = DIM*jnrH;
 449
 450             /* load j atom coordinates */
 451             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 452                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 453                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 454                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 455                                                  &jx0,&jy0,&jz0);
 456
 457             /* Calculate displacement vector */
 458             dx00             = _mm256_sub_ps(ix0,jx0);
 459             dy00             = _mm256_sub_ps(iy0,jy0);
 460             dz00             = _mm256_sub_ps(iz0,jz0);
 461             dx10             = _mm256_sub_ps(ix1,jx0);
 462             dy10             = _mm256_sub_ps(iy1,jy0);
 463             dz10             = _mm256_sub_ps(iz1,jz0);
 464             dx20             = _mm256_sub_ps(ix2,jx0);
 465             dy20             = _mm256_sub_ps(iy2,jy0);
 466             dz20             = _mm256_sub_ps(iz2,jz0);
 467             dx30             = _mm256_sub_ps(ix3,jx0);
 468             dy30             = _mm256_sub_ps(iy3,jy0);
 469             dz30             = _mm256_sub_ps(iz3,jz0);
 470
 471             /* Calculate squared distance and things based on it */
 472             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 473             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 474             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 475             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 476
 477             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 478             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 479             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 480
 481             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 482             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 483             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 484             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 485
 486             /* Load parameters for j particles */
 487             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 488                                                                  charge+jnrC+0,charge+jnrD+0,
 489                                                                  charge+jnrE+0,charge+jnrF+0,
 490                                                                  charge+jnrG+0,charge+jnrH+0);
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493             vdwjidx0C        = 2*vdwtype[jnrC+0];
 494             vdwjidx0D        = 2*vdwtype[jnrD+0];
 495             vdwjidx0E        = 2*vdwtype[jnrE+0];
 496             vdwjidx0F        = 2*vdwtype[jnrF+0];
 497             vdwjidx0G        = 2*vdwtype[jnrG+0];
 498             vdwjidx0H        = 2*vdwtype[jnrH+0];
 499
 500             fjx0             = _mm256_setzero_ps();
 501             fjy0             = _mm256_setzero_ps();
 502             fjz0             = _mm256_setzero_ps();
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 510                                             vdwioffsetptr0+vdwjidx0B,
 511                                             vdwioffsetptr0+vdwjidx0C,
 512                                             vdwioffsetptr0+vdwjidx0D,
 513                                             vdwioffsetptr0+vdwjidx0E,
 514                                             vdwioffsetptr0+vdwjidx0F,
 515                                             vdwioffsetptr0+vdwjidx0G,
 516                                             vdwioffsetptr0+vdwjidx0H,
 517                                             &c6_00,&c12_00);
 518
 519             /* LENNARD-JONES DISPERSION/REPULSION */
 520
 521             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 522             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 523             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 524             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 525             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 529             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 530
 531             fscal            = fvdw;
 532
 533             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 534
 535             /* Calculate temporary vectorial force */
 536             tx               = _mm256_mul_ps(fscal,dx00);
 537             ty               = _mm256_mul_ps(fscal,dy00);
 538             tz               = _mm256_mul_ps(fscal,dz00);
 539
 540             /* Update vectorial force */
 541             fix0             = _mm256_add_ps(fix0,tx);
 542             fiy0             = _mm256_add_ps(fiy0,ty);
 543             fiz0             = _mm256_add_ps(fiz0,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             qq10             = _mm256_mul_ps(iq1,jq0);
 555
 556             /* COULOMB ELECTROSTATICS */
 557             velec            = _mm256_mul_ps(qq10,rinv10);
 558             felec            = _mm256_mul_ps(velec,rinvsq10);
 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,dx10);
 570             ty               = _mm256_mul_ps(fscal,dy10);
 571             tz               = _mm256_mul_ps(fscal,dz10);
 572
 573             /* Update vectorial force */
 574             fix1             = _mm256_add_ps(fix1,tx);
 575             fiy1             = _mm256_add_ps(fiy1,ty);
 576             fiz1             = _mm256_add_ps(fiz1,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             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             /* Compute parameters for interactions between i and j atoms */
 587             qq20             = _mm256_mul_ps(iq2,jq0);
 588
 589             /* COULOMB ELECTROSTATICS */
 590             velec            = _mm256_mul_ps(qq20,rinv20);
 591             felec            = _mm256_mul_ps(velec,rinvsq20);
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm256_andnot_ps(dummy_mask,velec);
 595             velecsum         = _mm256_add_ps(velecsum,velec);
 596
 597             fscal            = felec;
 598
 599             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm256_mul_ps(fscal,dx20);
 603             ty               = _mm256_mul_ps(fscal,dy20);
 604             tz               = _mm256_mul_ps(fscal,dz20);
 605
 606             /* Update vectorial force */
 607             fix2             = _mm256_add_ps(fix2,tx);
 608             fiy2             = _mm256_add_ps(fiy2,ty);
 609             fiz2             = _mm256_add_ps(fiz2,tz);
 610
 611             fjx0             = _mm256_add_ps(fjx0,tx);
 612             fjy0             = _mm256_add_ps(fjy0,ty);
 613             fjz0             = _mm256_add_ps(fjz0,tz);
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             /* Compute parameters for interactions between i and j atoms */
 620             qq30             = _mm256_mul_ps(iq3,jq0);
 621
 622             /* COULOMB ELECTROSTATICS */
 623             velec            = _mm256_mul_ps(qq30,rinv30);
 624             felec            = _mm256_mul_ps(velec,rinvsq30);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm256_andnot_ps(dummy_mask,velec);
 628             velecsum         = _mm256_add_ps(velecsum,velec);
 629
 630             fscal            = felec;
 631
 632             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm256_mul_ps(fscal,dx30);
 636             ty               = _mm256_mul_ps(fscal,dy30);
 637             tz               = _mm256_mul_ps(fscal,dz30);
 638
 639             /* Update vectorial force */
 640             fix3             = _mm256_add_ps(fix3,tx);
 641             fiy3             = _mm256_add_ps(fiy3,ty);
 642             fiz3             = _mm256_add_ps(fiz3,tz);
 643
 644             fjx0             = _mm256_add_ps(fjx0,tx);
 645             fjy0             = _mm256_add_ps(fjy0,ty);
 646             fjz0             = _mm256_add_ps(fjz0,tz);
 647
 648             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 649             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 650             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 651             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 652             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 653             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 654             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 655             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 656
 657             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 658
 659             /* Inner loop uses 116 flops */
 660         }
 661
 662         /* End of innermost loop */
 663
 664         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 665                                                  f+i_coord_offset,fshift+i_shift_offset);
 666
 667         ggid                        = gid[iidx];
 668         /* Update potential energies */
 669         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 670         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 671
 672         /* Increment number of inner iterations */
 673         inneriter                  += j_index_end - j_index_start;
 674
 675         /* Outer loop uses 26 flops */
 676     }
 677
 678     /* Increment number of outer iterations */
 679     outeriter        += nri;
 680
 681     /* Update outer/inner flops */
 682
 683     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*116);
 684 }
 685 /*
 686  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_single
 687  * Electrostatics interaction: Coulomb
 688  * VdW interaction:            LennardJones
 689  * Geometry:                   Water4-Particle
 690  * Calculate force/pot:        Force
 691  */
 692 void
 693 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_single
 694                     (t_nblist                    * gmx_restrict       nlist,
 695                      rvec                        * gmx_restrict          xx,
 696                      rvec                        * gmx_restrict          ff,
 697                      t_forcerec                  * gmx_restrict          fr,
 698                      t_mdatoms                   * gmx_restrict     mdatoms,
 699                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 700                      t_nrnb                      * gmx_restrict        nrnb)
 701 {
 702     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 703      * just 0 for non-waters.
 704      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 705      * jnr indices corresponding to data put in the four positions in the SIMD register.
 706      */
 707     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 708     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 709     int              jnrA,jnrB,jnrC,jnrD;
 710     int              jnrE,jnrF,jnrG,jnrH;
 711     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 712     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 713     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 714     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 715     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 716     real             rcutoff_scalar;
 717     real             *shiftvec,*fshift,*x,*f;
 718     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 719     real             scratch[4*DIM];
 720     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 721     real *           vdwioffsetptr0;
 722     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 723     real *           vdwioffsetptr1;
 724     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 725     real *           vdwioffsetptr2;
 726     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 727     real *           vdwioffsetptr3;
 728     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 729     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 730     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 731     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 732     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 733     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 734     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 735     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 736     real             *charge;
 737     int              nvdwtype;
 738     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 739     int              *vdwtype;
 740     real             *vdwparam;
 741     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 742     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 743     __m256           dummy_mask,cutoff_mask;
 744     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 745     __m256           one     = _mm256_set1_ps(1.0);
 746     __m256           two     = _mm256_set1_ps(2.0);
 747     x                = xx[0];
 748     f                = ff[0];
 749
 750     nri              = nlist->nri;
 751     iinr             = nlist->iinr;
 752     jindex           = nlist->jindex;
 753     jjnr             = nlist->jjnr;
 754     shiftidx         = nlist->shift;
 755     gid              = nlist->gid;
 756     shiftvec         = fr->shift_vec[0];
 757     fshift           = fr->fshift[0];
 758     facel            = _mm256_set1_ps(fr->epsfac);
 759     charge           = mdatoms->chargeA;
 760     nvdwtype         = fr->ntype;
 761     vdwparam         = fr->nbfp;
 762     vdwtype          = mdatoms->typeA;
 763
 764     /* Setup water-specific parameters */
 765     inr              = nlist->iinr[0];
 766     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 767     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 768     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 769     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 770
 771     /* Avoid stupid compiler warnings */
 772     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 773     j_coord_offsetA = 0;
 774     j_coord_offsetB = 0;
 775     j_coord_offsetC = 0;
 776     j_coord_offsetD = 0;
 777     j_coord_offsetE = 0;
 778     j_coord_offsetF = 0;
 779     j_coord_offsetG = 0;
 780     j_coord_offsetH = 0;
 781
 782     outeriter        = 0;
 783     inneriter        = 0;
 784
 785     for(iidx=0;iidx<4*DIM;iidx++)
 786     {
 787         scratch[iidx] = 0.0;
 788     }
 789
 790     /* Start outer loop over neighborlists */
 791     for(iidx=0; iidx<nri; iidx++)
 792     {
 793         /* Load shift vector for this list */
 794         i_shift_offset   = DIM*shiftidx[iidx];
 795
 796         /* Load limits for loop over neighbors */
 797         j_index_start    = jindex[iidx];
 798         j_index_end      = jindex[iidx+1];
 799
 800         /* Get outer coordinate index */
 801         inr              = iinr[iidx];
 802         i_coord_offset   = DIM*inr;
 803
 804         /* Load i particle coords and add shift vector */
 805         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 806                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 807
 808         fix0             = _mm256_setzero_ps();
 809         fiy0             = _mm256_setzero_ps();
 810         fiz0             = _mm256_setzero_ps();
 811         fix1             = _mm256_setzero_ps();
 812         fiy1             = _mm256_setzero_ps();
 813         fiz1             = _mm256_setzero_ps();
 814         fix2             = _mm256_setzero_ps();
 815         fiy2             = _mm256_setzero_ps();
 816         fiz2             = _mm256_setzero_ps();
 817         fix3             = _mm256_setzero_ps();
 818         fiy3             = _mm256_setzero_ps();
 819         fiz3             = _mm256_setzero_ps();
 820
 821         /* Start inner kernel loop */
 822         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 823         {
 824
 825             /* Get j neighbor index, and coordinate index */
 826             jnrA             = jjnr[jidx];
 827             jnrB             = jjnr[jidx+1];
 828             jnrC             = jjnr[jidx+2];
 829             jnrD             = jjnr[jidx+3];
 830             jnrE             = jjnr[jidx+4];
 831             jnrF             = jjnr[jidx+5];
 832             jnrG             = jjnr[jidx+6];
 833             jnrH             = jjnr[jidx+7];
 834             j_coord_offsetA  = DIM*jnrA;
 835             j_coord_offsetB  = DIM*jnrB;
 836             j_coord_offsetC  = DIM*jnrC;
 837             j_coord_offsetD  = DIM*jnrD;
 838             j_coord_offsetE  = DIM*jnrE;
 839             j_coord_offsetF  = DIM*jnrF;
 840             j_coord_offsetG  = DIM*jnrG;
 841             j_coord_offsetH  = DIM*jnrH;
 842
 843             /* load j atom coordinates */
 844             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 845                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 846                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 847                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 848                                                  &jx0,&jy0,&jz0);
 849
 850             /* Calculate displacement vector */
 851             dx00             = _mm256_sub_ps(ix0,jx0);
 852             dy00             = _mm256_sub_ps(iy0,jy0);
 853             dz00             = _mm256_sub_ps(iz0,jz0);
 854             dx10             = _mm256_sub_ps(ix1,jx0);
 855             dy10             = _mm256_sub_ps(iy1,jy0);
 856             dz10             = _mm256_sub_ps(iz1,jz0);
 857             dx20             = _mm256_sub_ps(ix2,jx0);
 858             dy20             = _mm256_sub_ps(iy2,jy0);
 859             dz20             = _mm256_sub_ps(iz2,jz0);
 860             dx30             = _mm256_sub_ps(ix3,jx0);
 861             dy30             = _mm256_sub_ps(iy3,jy0);
 862             dz30             = _mm256_sub_ps(iz3,jz0);
 863
 864             /* Calculate squared distance and things based on it */
 865             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 866             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 867             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 868             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 869
 870             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 871             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 872             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 873
 874             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 875             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 876             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 877             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 878
 879             /* Load parameters for j particles */
 880             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 881                                                                  charge+jnrC+0,charge+jnrD+0,
 882                                                                  charge+jnrE+0,charge+jnrF+0,
 883                                                                  charge+jnrG+0,charge+jnrH+0);
 884             vdwjidx0A        = 2*vdwtype[jnrA+0];
 885             vdwjidx0B        = 2*vdwtype[jnrB+0];
 886             vdwjidx0C        = 2*vdwtype[jnrC+0];
 887             vdwjidx0D        = 2*vdwtype[jnrD+0];
 888             vdwjidx0E        = 2*vdwtype[jnrE+0];
 889             vdwjidx0F        = 2*vdwtype[jnrF+0];
 890             vdwjidx0G        = 2*vdwtype[jnrG+0];
 891             vdwjidx0H        = 2*vdwtype[jnrH+0];
 892
 893             fjx0             = _mm256_setzero_ps();
 894             fjy0             = _mm256_setzero_ps();
 895             fjz0             = _mm256_setzero_ps();
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 903                                             vdwioffsetptr0+vdwjidx0B,
 904                                             vdwioffsetptr0+vdwjidx0C,
 905                                             vdwioffsetptr0+vdwjidx0D,
 906                                             vdwioffsetptr0+vdwjidx0E,
 907                                             vdwioffsetptr0+vdwjidx0F,
 908                                             vdwioffsetptr0+vdwjidx0G,
 909                                             vdwioffsetptr0+vdwjidx0H,
 910                                             &c6_00,&c12_00);
 911
 912             /* LENNARD-JONES DISPERSION/REPULSION */
 913
 914             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 915             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 916
 917             fscal            = fvdw;
 918
 919             /* Calculate temporary vectorial force */
 920             tx               = _mm256_mul_ps(fscal,dx00);
 921             ty               = _mm256_mul_ps(fscal,dy00);
 922             tz               = _mm256_mul_ps(fscal,dz00);
 923
 924             /* Update vectorial force */
 925             fix0             = _mm256_add_ps(fix0,tx);
 926             fiy0             = _mm256_add_ps(fiy0,ty);
 927             fiz0             = _mm256_add_ps(fiz0,tz);
 928
 929             fjx0             = _mm256_add_ps(fjx0,tx);
 930             fjy0             = _mm256_add_ps(fjy0,ty);
 931             fjz0             = _mm256_add_ps(fjz0,tz);
 932
 933             /**************************
 934              * CALCULATE INTERACTIONS *
 935              **************************/
 936
 937             /* Compute parameters for interactions between i and j atoms */
 938             qq10             = _mm256_mul_ps(iq1,jq0);
 939
 940             /* COULOMB ELECTROSTATICS */
 941             velec            = _mm256_mul_ps(qq10,rinv10);
 942             felec            = _mm256_mul_ps(velec,rinvsq10);
 943
 944             fscal            = felec;
 945
 946             /* Calculate temporary vectorial force */
 947             tx               = _mm256_mul_ps(fscal,dx10);
 948             ty               = _mm256_mul_ps(fscal,dy10);
 949             tz               = _mm256_mul_ps(fscal,dz10);
 950
 951             /* Update vectorial force */
 952             fix1             = _mm256_add_ps(fix1,tx);
 953             fiy1             = _mm256_add_ps(fiy1,ty);
 954             fiz1             = _mm256_add_ps(fiz1,tz);
 955
 956             fjx0             = _mm256_add_ps(fjx0,tx);
 957             fjy0             = _mm256_add_ps(fjy0,ty);
 958             fjz0             = _mm256_add_ps(fjz0,tz);
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq20             = _mm256_mul_ps(iq2,jq0);
 966
 967             /* COULOMB ELECTROSTATICS */
 968             velec            = _mm256_mul_ps(qq20,rinv20);
 969             felec            = _mm256_mul_ps(velec,rinvsq20);
 970
 971             fscal            = felec;
 972
 973             /* Calculate temporary vectorial force */
 974             tx               = _mm256_mul_ps(fscal,dx20);
 975             ty               = _mm256_mul_ps(fscal,dy20);
 976             tz               = _mm256_mul_ps(fscal,dz20);
 977
 978             /* Update vectorial force */
 979             fix2             = _mm256_add_ps(fix2,tx);
 980             fiy2             = _mm256_add_ps(fiy2,ty);
 981             fiz2             = _mm256_add_ps(fiz2,tz);
 982
 983             fjx0             = _mm256_add_ps(fjx0,tx);
 984             fjy0             = _mm256_add_ps(fjy0,ty);
 985             fjz0             = _mm256_add_ps(fjz0,tz);
 986
 987             /**************************
 988              * CALCULATE INTERACTIONS *
 989              **************************/
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq30             = _mm256_mul_ps(iq3,jq0);
 993
 994             /* COULOMB ELECTROSTATICS */
 995             velec            = _mm256_mul_ps(qq30,rinv30);
 996             felec            = _mm256_mul_ps(velec,rinvsq30);
 997
 998             fscal            = felec;
 999
1000             /* Calculate temporary vectorial force */
1001             tx               = _mm256_mul_ps(fscal,dx30);
1002             ty               = _mm256_mul_ps(fscal,dy30);
1003             tz               = _mm256_mul_ps(fscal,dz30);
1004
1005             /* Update vectorial force */
1006             fix3             = _mm256_add_ps(fix3,tx);
1007             fiy3             = _mm256_add_ps(fiy3,ty);
1008             fiz3             = _mm256_add_ps(fiz3,tz);
1009
1010             fjx0             = _mm256_add_ps(fjx0,tx);
1011             fjy0             = _mm256_add_ps(fjy0,ty);
1012             fjz0             = _mm256_add_ps(fjz0,tz);
1013
1014             fjptrA             = f+j_coord_offsetA;
1015             fjptrB             = f+j_coord_offsetB;
1016             fjptrC             = f+j_coord_offsetC;
1017             fjptrD             = f+j_coord_offsetD;
1018             fjptrE             = f+j_coord_offsetE;
1019             fjptrF             = f+j_coord_offsetF;
1020             fjptrG             = f+j_coord_offsetG;
1021             fjptrH             = f+j_coord_offsetH;
1022
1023             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1024
1025             /* Inner loop uses 108 flops */
1026         }
1027
1028         if(jidx<j_index_end)
1029         {
1030
1031             /* Get j neighbor index, and coordinate index */
1032             jnrlistA         = jjnr[jidx];
1033             jnrlistB         = jjnr[jidx+1];
1034             jnrlistC         = jjnr[jidx+2];
1035             jnrlistD         = jjnr[jidx+3];
1036             jnrlistE         = jjnr[jidx+4];
1037             jnrlistF         = jjnr[jidx+5];
1038             jnrlistG         = jjnr[jidx+6];
1039             jnrlistH         = jjnr[jidx+7];
1040             /* Sign of each element will be negative for non-real atoms.
1041              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1042              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1043              */
1044             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1045                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1046
1047             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1048             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1049             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1050             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1051             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1052             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1053             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1054             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1055             j_coord_offsetA  = DIM*jnrA;
1056             j_coord_offsetB  = DIM*jnrB;
1057             j_coord_offsetC  = DIM*jnrC;
1058             j_coord_offsetD  = DIM*jnrD;
1059             j_coord_offsetE  = DIM*jnrE;
1060             j_coord_offsetF  = DIM*jnrF;
1061             j_coord_offsetG  = DIM*jnrG;
1062             j_coord_offsetH  = DIM*jnrH;
1063
1064             /* load j atom coordinates */
1065             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1066                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1067                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1068                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1069                                                  &jx0,&jy0,&jz0);
1070
1071             /* Calculate displacement vector */
1072             dx00             = _mm256_sub_ps(ix0,jx0);
1073             dy00             = _mm256_sub_ps(iy0,jy0);
1074             dz00             = _mm256_sub_ps(iz0,jz0);
1075             dx10             = _mm256_sub_ps(ix1,jx0);
1076             dy10             = _mm256_sub_ps(iy1,jy0);
1077             dz10             = _mm256_sub_ps(iz1,jz0);
1078             dx20             = _mm256_sub_ps(ix2,jx0);
1079             dy20             = _mm256_sub_ps(iy2,jy0);
1080             dz20             = _mm256_sub_ps(iz2,jz0);
1081             dx30             = _mm256_sub_ps(ix3,jx0);
1082             dy30             = _mm256_sub_ps(iy3,jy0);
1083             dz30             = _mm256_sub_ps(iz3,jz0);
1084
1085             /* Calculate squared distance and things based on it */
1086             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1087             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1088             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1089             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1090
1091             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1092             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1093             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1094
1095             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1096             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1097             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1098             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1099
1100             /* Load parameters for j particles */
1101             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1102                                                                  charge+jnrC+0,charge+jnrD+0,
1103                                                                  charge+jnrE+0,charge+jnrF+0,
1104                                                                  charge+jnrG+0,charge+jnrH+0);
1105             vdwjidx0A        = 2*vdwtype[jnrA+0];
1106             vdwjidx0B        = 2*vdwtype[jnrB+0];
1107             vdwjidx0C        = 2*vdwtype[jnrC+0];
1108             vdwjidx0D        = 2*vdwtype[jnrD+0];
1109             vdwjidx0E        = 2*vdwtype[jnrE+0];
1110             vdwjidx0F        = 2*vdwtype[jnrF+0];
1111             vdwjidx0G        = 2*vdwtype[jnrG+0];
1112             vdwjidx0H        = 2*vdwtype[jnrH+0];
1113
1114             fjx0             = _mm256_setzero_ps();
1115             fjy0             = _mm256_setzero_ps();
1116             fjz0             = _mm256_setzero_ps();
1117
1118             /**************************
1119              * CALCULATE INTERACTIONS *
1120              **************************/
1121
1122             /* Compute parameters for interactions between i and j atoms */
1123             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1124                                             vdwioffsetptr0+vdwjidx0B,
1125                                             vdwioffsetptr0+vdwjidx0C,
1126                                             vdwioffsetptr0+vdwjidx0D,
1127                                             vdwioffsetptr0+vdwjidx0E,
1128                                             vdwioffsetptr0+vdwjidx0F,
1129                                             vdwioffsetptr0+vdwjidx0G,
1130                                             vdwioffsetptr0+vdwjidx0H,
1131                                             &c6_00,&c12_00);
1132
1133             /* LENNARD-JONES DISPERSION/REPULSION */
1134
1135             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1136             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1137
1138             fscal            = fvdw;
1139
1140             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1141
1142             /* Calculate temporary vectorial force */
1143             tx               = _mm256_mul_ps(fscal,dx00);
1144             ty               = _mm256_mul_ps(fscal,dy00);
1145             tz               = _mm256_mul_ps(fscal,dz00);
1146
1147             /* Update vectorial force */
1148             fix0             = _mm256_add_ps(fix0,tx);
1149             fiy0             = _mm256_add_ps(fiy0,ty);
1150             fiz0             = _mm256_add_ps(fiz0,tz);
1151
1152             fjx0             = _mm256_add_ps(fjx0,tx);
1153             fjy0             = _mm256_add_ps(fjy0,ty);
1154             fjz0             = _mm256_add_ps(fjz0,tz);
1155
1156             /**************************
1157              * CALCULATE INTERACTIONS *
1158              **************************/
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             qq10             = _mm256_mul_ps(iq1,jq0);
1162
1163             /* COULOMB ELECTROSTATICS */
1164             velec            = _mm256_mul_ps(qq10,rinv10);
1165             felec            = _mm256_mul_ps(velec,rinvsq10);
1166
1167             fscal            = felec;
1168
1169             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1170
1171             /* Calculate temporary vectorial force */
1172             tx               = _mm256_mul_ps(fscal,dx10);
1173             ty               = _mm256_mul_ps(fscal,dy10);
1174             tz               = _mm256_mul_ps(fscal,dz10);
1175
1176             /* Update vectorial force */
1177             fix1             = _mm256_add_ps(fix1,tx);
1178             fiy1             = _mm256_add_ps(fiy1,ty);
1179             fiz1             = _mm256_add_ps(fiz1,tz);
1180
1181             fjx0             = _mm256_add_ps(fjx0,tx);
1182             fjy0             = _mm256_add_ps(fjy0,ty);
1183             fjz0             = _mm256_add_ps(fjz0,tz);
1184
1185             /**************************
1186              * CALCULATE INTERACTIONS *
1187              **************************/
1188
1189             /* Compute parameters for interactions between i and j atoms */
1190             qq20             = _mm256_mul_ps(iq2,jq0);
1191
1192             /* COULOMB ELECTROSTATICS */
1193             velec            = _mm256_mul_ps(qq20,rinv20);
1194             felec            = _mm256_mul_ps(velec,rinvsq20);
1195
1196             fscal            = felec;
1197
1198             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1199
1200             /* Calculate temporary vectorial force */
1201             tx               = _mm256_mul_ps(fscal,dx20);
1202             ty               = _mm256_mul_ps(fscal,dy20);
1203             tz               = _mm256_mul_ps(fscal,dz20);
1204
1205             /* Update vectorial force */
1206             fix2             = _mm256_add_ps(fix2,tx);
1207             fiy2             = _mm256_add_ps(fiy2,ty);
1208             fiz2             = _mm256_add_ps(fiz2,tz);
1209
1210             fjx0             = _mm256_add_ps(fjx0,tx);
1211             fjy0             = _mm256_add_ps(fjy0,ty);
1212             fjz0             = _mm256_add_ps(fjz0,tz);
1213
1214             /**************************
1215              * CALCULATE INTERACTIONS *
1216              **************************/
1217
1218             /* Compute parameters for interactions between i and j atoms */
1219             qq30             = _mm256_mul_ps(iq3,jq0);
1220
1221             /* COULOMB ELECTROSTATICS */
1222             velec            = _mm256_mul_ps(qq30,rinv30);
1223             felec            = _mm256_mul_ps(velec,rinvsq30);
1224
1225             fscal            = felec;
1226
1227             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1228
1229             /* Calculate temporary vectorial force */
1230             tx               = _mm256_mul_ps(fscal,dx30);
1231             ty               = _mm256_mul_ps(fscal,dy30);
1232             tz               = _mm256_mul_ps(fscal,dz30);
1233
1234             /* Update vectorial force */
1235             fix3             = _mm256_add_ps(fix3,tx);
1236             fiy3             = _mm256_add_ps(fiy3,ty);
1237             fiz3             = _mm256_add_ps(fiz3,tz);
1238
1239             fjx0             = _mm256_add_ps(fjx0,tx);
1240             fjy0             = _mm256_add_ps(fjy0,ty);
1241             fjz0             = _mm256_add_ps(fjz0,tz);
1242
1243             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1244             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1245             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1246             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1247             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1248             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1249             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1250             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1251
1252             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1253
1254             /* Inner loop uses 108 flops */
1255         }
1256
1257         /* End of innermost loop */
1258
1259         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1260                                                  f+i_coord_offset,fshift+i_shift_offset);
1261
1262         /* Increment number of inner iterations */
1263         inneriter                  += j_index_end - j_index_start;
1264
1265         /* Outer loop uses 24 flops */
1266     }
1267
1268     /* Increment number of outer iterations */
1269     outeriter        += nri;
1270
1271     /* Update outer/inner flops */
1272
1273     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1274 }