src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwLJ_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_avx_256_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  88     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  97     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  98     __m256           dummy_mask,cutoff_mask;
  99     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 100     __m256           one     = _mm256_set1_ps(1.0);
 101     __m256           two     = _mm256_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_ps(fr->ic->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125     j_coord_offsetE = 0;
 126     j_coord_offsetF = 0;
 127     j_coord_offsetG = 0;
 128     j_coord_offsetH = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm256_setzero_ps();
 156         fiy0             = _mm256_setzero_ps();
 157         fiz0             = _mm256_setzero_ps();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 161         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm256_setzero_ps();
 165         vvdwsum          = _mm256_setzero_ps();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             jnrC             = jjnr[jidx+2];
 175             jnrD             = jjnr[jidx+3];
 176             jnrE             = jjnr[jidx+4];
 177             jnrF             = jjnr[jidx+5];
 178             jnrG             = jjnr[jidx+6];
 179             jnrH             = jjnr[jidx+7];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184             j_coord_offsetE  = DIM*jnrE;
 185             j_coord_offsetF  = DIM*jnrF;
 186             j_coord_offsetG  = DIM*jnrG;
 187             j_coord_offsetH  = DIM*jnrH;
 188
 189             /* load j atom coordinates */
 190             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 192                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 193                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 194                                                  &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm256_sub_ps(ix0,jx0);
 198             dy00             = _mm256_sub_ps(iy0,jy0);
 199             dz00             = _mm256_sub_ps(iz0,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 203
 204             rinv00           = avx256_invsqrt_f(rsq00);
 205
 206             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 207
 208             /* Load parameters for j particles */
 209             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 210                                                                  charge+jnrC+0,charge+jnrD+0,
 211                                                                  charge+jnrE+0,charge+jnrF+0,
 212                                                                  charge+jnrG+0,charge+jnrH+0);
 213             vdwjidx0A        = 2*vdwtype[jnrA+0];
 214             vdwjidx0B        = 2*vdwtype[jnrB+0];
 215             vdwjidx0C        = 2*vdwtype[jnrC+0];
 216             vdwjidx0D        = 2*vdwtype[jnrD+0];
 217             vdwjidx0E        = 2*vdwtype[jnrE+0];
 218             vdwjidx0F        = 2*vdwtype[jnrF+0];
 219             vdwjidx0G        = 2*vdwtype[jnrG+0];
 220             vdwjidx0H        = 2*vdwtype[jnrH+0];
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq00             = _mm256_mul_ps(iq0,jq0);
 228             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 229                                             vdwioffsetptr0+vdwjidx0B,
 230                                             vdwioffsetptr0+vdwjidx0C,
 231                                             vdwioffsetptr0+vdwjidx0D,
 232                                             vdwioffsetptr0+vdwjidx0E,
 233                                             vdwioffsetptr0+vdwjidx0F,
 234                                             vdwioffsetptr0+vdwjidx0G,
 235                                             vdwioffsetptr0+vdwjidx0H,
 236                                             &c6_00,&c12_00);
 237
 238             /* COULOMB ELECTROSTATICS */
 239             velec            = _mm256_mul_ps(qq00,rinv00);
 240             felec            = _mm256_mul_ps(velec,rinvsq00);
 241
 242             /* LENNARD-JONES DISPERSION/REPULSION */
 243
 244             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 245             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 246             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 247             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 248             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 249
 250             /* Update potential sum for this i atom from the interaction with this j atom. */
 251             velecsum         = _mm256_add_ps(velecsum,velec);
 252             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 253
 254             fscal            = _mm256_add_ps(felec,fvdw);
 255
 256             /* Calculate temporary vectorial force */
 257             tx               = _mm256_mul_ps(fscal,dx00);
 258             ty               = _mm256_mul_ps(fscal,dy00);
 259             tz               = _mm256_mul_ps(fscal,dz00);
 260
 261             /* Update vectorial force */
 262             fix0             = _mm256_add_ps(fix0,tx);
 263             fiy0             = _mm256_add_ps(fiy0,ty);
 264             fiz0             = _mm256_add_ps(fiz0,tz);
 265
 266             fjptrA             = f+j_coord_offsetA;
 267             fjptrB             = f+j_coord_offsetB;
 268             fjptrC             = f+j_coord_offsetC;
 269             fjptrD             = f+j_coord_offsetD;
 270             fjptrE             = f+j_coord_offsetE;
 271             fjptrF             = f+j_coord_offsetF;
 272             fjptrG             = f+j_coord_offsetG;
 273             fjptrH             = f+j_coord_offsetH;
 274             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 275
 276             /* Inner loop uses 39 flops */
 277         }
 278
 279         if(jidx<j_index_end)
 280         {
 281
 282             /* Get j neighbor index, and coordinate index */
 283             jnrlistA         = jjnr[jidx];
 284             jnrlistB         = jjnr[jidx+1];
 285             jnrlistC         = jjnr[jidx+2];
 286             jnrlistD         = jjnr[jidx+3];
 287             jnrlistE         = jjnr[jidx+4];
 288             jnrlistF         = jjnr[jidx+5];
 289             jnrlistG         = jjnr[jidx+6];
 290             jnrlistH         = jjnr[jidx+7];
 291             /* Sign of each element will be negative for non-real atoms.
 292              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 293              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 294              */
 295             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 296                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 297
 298             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 299             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 300             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 301             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 302             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 303             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 304             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 305             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 306             j_coord_offsetA  = DIM*jnrA;
 307             j_coord_offsetB  = DIM*jnrB;
 308             j_coord_offsetC  = DIM*jnrC;
 309             j_coord_offsetD  = DIM*jnrD;
 310             j_coord_offsetE  = DIM*jnrE;
 311             j_coord_offsetF  = DIM*jnrF;
 312             j_coord_offsetG  = DIM*jnrG;
 313             j_coord_offsetH  = DIM*jnrH;
 314
 315             /* load j atom coordinates */
 316             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 317                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 318                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 319                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 320                                                  &jx0,&jy0,&jz0);
 321
 322             /* Calculate displacement vector */
 323             dx00             = _mm256_sub_ps(ix0,jx0);
 324             dy00             = _mm256_sub_ps(iy0,jy0);
 325             dz00             = _mm256_sub_ps(iz0,jz0);
 326
 327             /* Calculate squared distance and things based on it */
 328             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 329
 330             rinv00           = avx256_invsqrt_f(rsq00);
 331
 332             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 333
 334             /* Load parameters for j particles */
 335             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 336                                                                  charge+jnrC+0,charge+jnrD+0,
 337                                                                  charge+jnrE+0,charge+jnrF+0,
 338                                                                  charge+jnrG+0,charge+jnrH+0);
 339             vdwjidx0A        = 2*vdwtype[jnrA+0];
 340             vdwjidx0B        = 2*vdwtype[jnrB+0];
 341             vdwjidx0C        = 2*vdwtype[jnrC+0];
 342             vdwjidx0D        = 2*vdwtype[jnrD+0];
 343             vdwjidx0E        = 2*vdwtype[jnrE+0];
 344             vdwjidx0F        = 2*vdwtype[jnrF+0];
 345             vdwjidx0G        = 2*vdwtype[jnrG+0];
 346             vdwjidx0H        = 2*vdwtype[jnrH+0];
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq00             = _mm256_mul_ps(iq0,jq0);
 354             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 355                                             vdwioffsetptr0+vdwjidx0B,
 356                                             vdwioffsetptr0+vdwjidx0C,
 357                                             vdwioffsetptr0+vdwjidx0D,
 358                                             vdwioffsetptr0+vdwjidx0E,
 359                                             vdwioffsetptr0+vdwjidx0F,
 360                                             vdwioffsetptr0+vdwjidx0G,
 361                                             vdwioffsetptr0+vdwjidx0H,
 362                                             &c6_00,&c12_00);
 363
 364             /* COULOMB ELECTROSTATICS */
 365             velec            = _mm256_mul_ps(qq00,rinv00);
 366             felec            = _mm256_mul_ps(velec,rinvsq00);
 367
 368             /* LENNARD-JONES DISPERSION/REPULSION */
 369
 370             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 371             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 372             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 373             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 374             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velec            = _mm256_andnot_ps(dummy_mask,velec);
 378             velecsum         = _mm256_add_ps(velecsum,velec);
 379             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 380             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 381
 382             fscal            = _mm256_add_ps(felec,fvdw);
 383
 384             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 385
 386             /* Calculate temporary vectorial force */
 387             tx               = _mm256_mul_ps(fscal,dx00);
 388             ty               = _mm256_mul_ps(fscal,dy00);
 389             tz               = _mm256_mul_ps(fscal,dz00);
 390
 391             /* Update vectorial force */
 392             fix0             = _mm256_add_ps(fix0,tx);
 393             fiy0             = _mm256_add_ps(fiy0,ty);
 394             fiz0             = _mm256_add_ps(fiz0,tz);
 395
 396             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 397             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 398             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 399             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 400             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 401             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 402             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 403             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 404             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 405
 406             /* Inner loop uses 39 flops */
 407         }
 408
 409         /* End of innermost loop */
 410
 411         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 412                                                  f+i_coord_offset,fshift+i_shift_offset);
 413
 414         ggid                        = gid[iidx];
 415         /* Update potential energies */
 416         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 417         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 418
 419         /* Increment number of inner iterations */
 420         inneriter                  += j_index_end - j_index_start;
 421
 422         /* Outer loop uses 9 flops */
 423     }
 424
 425     /* Increment number of outer iterations */
 426     outeriter        += nri;
 427
 428     /* Update outer/inner flops */
 429
 430     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*39);
 431 }
 432 /*
 433  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_single
 434  * Electrostatics interaction: Coulomb
 435  * VdW interaction:            LennardJones
 436  * Geometry:                   Particle-Particle
 437  * Calculate force/pot:        Force
 438  */
 439 void
 440 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_single
 441                     (t_nblist                    * gmx_restrict       nlist,
 442                      rvec                        * gmx_restrict          xx,
 443                      rvec                        * gmx_restrict          ff,
 444                      struct t_forcerec           * gmx_restrict          fr,
 445                      t_mdatoms                   * gmx_restrict     mdatoms,
 446                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 447                      t_nrnb                      * gmx_restrict        nrnb)
 448 {
 449     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 450      * just 0 for non-waters.
 451      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 452      * jnr indices corresponding to data put in the four positions in the SIMD register.
 453      */
 454     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 455     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 456     int              jnrA,jnrB,jnrC,jnrD;
 457     int              jnrE,jnrF,jnrG,jnrH;
 458     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 459     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 460     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 461     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 462     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 463     real             rcutoff_scalar;
 464     real             *shiftvec,*fshift,*x,*f;
 465     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 466     real             scratch[4*DIM];
 467     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 468     real *           vdwioffsetptr0;
 469     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 470     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 471     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 472     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 473     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 474     real             *charge;
 475     int              nvdwtype;
 476     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 477     int              *vdwtype;
 478     real             *vdwparam;
 479     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 480     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 481     __m256           dummy_mask,cutoff_mask;
 482     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 483     __m256           one     = _mm256_set1_ps(1.0);
 484     __m256           two     = _mm256_set1_ps(2.0);
 485     x                = xx[0];
 486     f                = ff[0];
 487
 488     nri              = nlist->nri;
 489     iinr             = nlist->iinr;
 490     jindex           = nlist->jindex;
 491     jjnr             = nlist->jjnr;
 492     shiftidx         = nlist->shift;
 493     gid              = nlist->gid;
 494     shiftvec         = fr->shift_vec[0];
 495     fshift           = fr->fshift[0];
 496     facel            = _mm256_set1_ps(fr->ic->epsfac);
 497     charge           = mdatoms->chargeA;
 498     nvdwtype         = fr->ntype;
 499     vdwparam         = fr->nbfp;
 500     vdwtype          = mdatoms->typeA;
 501
 502     /* Avoid stupid compiler warnings */
 503     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 504     j_coord_offsetA = 0;
 505     j_coord_offsetB = 0;
 506     j_coord_offsetC = 0;
 507     j_coord_offsetD = 0;
 508     j_coord_offsetE = 0;
 509     j_coord_offsetF = 0;
 510     j_coord_offsetG = 0;
 511     j_coord_offsetH = 0;
 512
 513     outeriter        = 0;
 514     inneriter        = 0;
 515
 516     for(iidx=0;iidx<4*DIM;iidx++)
 517     {
 518         scratch[iidx] = 0.0;
 519     }
 520
 521     /* Start outer loop over neighborlists */
 522     for(iidx=0; iidx<nri; iidx++)
 523     {
 524         /* Load shift vector for this list */
 525         i_shift_offset   = DIM*shiftidx[iidx];
 526
 527         /* Load limits for loop over neighbors */
 528         j_index_start    = jindex[iidx];
 529         j_index_end      = jindex[iidx+1];
 530
 531         /* Get outer coordinate index */
 532         inr              = iinr[iidx];
 533         i_coord_offset   = DIM*inr;
 534
 535         /* Load i particle coords and add shift vector */
 536         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 537
 538         fix0             = _mm256_setzero_ps();
 539         fiy0             = _mm256_setzero_ps();
 540         fiz0             = _mm256_setzero_ps();
 541
 542         /* Load parameters for i particles */
 543         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 544         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 545
 546         /* Start inner kernel loop */
 547         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 548         {
 549
 550             /* Get j neighbor index, and coordinate index */
 551             jnrA             = jjnr[jidx];
 552             jnrB             = jjnr[jidx+1];
 553             jnrC             = jjnr[jidx+2];
 554             jnrD             = jjnr[jidx+3];
 555             jnrE             = jjnr[jidx+4];
 556             jnrF             = jjnr[jidx+5];
 557             jnrG             = jjnr[jidx+6];
 558             jnrH             = jjnr[jidx+7];
 559             j_coord_offsetA  = DIM*jnrA;
 560             j_coord_offsetB  = DIM*jnrB;
 561             j_coord_offsetC  = DIM*jnrC;
 562             j_coord_offsetD  = DIM*jnrD;
 563             j_coord_offsetE  = DIM*jnrE;
 564             j_coord_offsetF  = DIM*jnrF;
 565             j_coord_offsetG  = DIM*jnrG;
 566             j_coord_offsetH  = DIM*jnrH;
 567
 568             /* load j atom coordinates */
 569             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 570                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 571                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 572                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 573                                                  &jx0,&jy0,&jz0);
 574
 575             /* Calculate displacement vector */
 576             dx00             = _mm256_sub_ps(ix0,jx0);
 577             dy00             = _mm256_sub_ps(iy0,jy0);
 578             dz00             = _mm256_sub_ps(iz0,jz0);
 579
 580             /* Calculate squared distance and things based on it */
 581             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 582
 583             rinv00           = avx256_invsqrt_f(rsq00);
 584
 585             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 586
 587             /* Load parameters for j particles */
 588             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 589                                                                  charge+jnrC+0,charge+jnrD+0,
 590                                                                  charge+jnrE+0,charge+jnrF+0,
 591                                                                  charge+jnrG+0,charge+jnrH+0);
 592             vdwjidx0A        = 2*vdwtype[jnrA+0];
 593             vdwjidx0B        = 2*vdwtype[jnrB+0];
 594             vdwjidx0C        = 2*vdwtype[jnrC+0];
 595             vdwjidx0D        = 2*vdwtype[jnrD+0];
 596             vdwjidx0E        = 2*vdwtype[jnrE+0];
 597             vdwjidx0F        = 2*vdwtype[jnrF+0];
 598             vdwjidx0G        = 2*vdwtype[jnrG+0];
 599             vdwjidx0H        = 2*vdwtype[jnrH+0];
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq00             = _mm256_mul_ps(iq0,jq0);
 607             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 608                                             vdwioffsetptr0+vdwjidx0B,
 609                                             vdwioffsetptr0+vdwjidx0C,
 610                                             vdwioffsetptr0+vdwjidx0D,
 611                                             vdwioffsetptr0+vdwjidx0E,
 612                                             vdwioffsetptr0+vdwjidx0F,
 613                                             vdwioffsetptr0+vdwjidx0G,
 614                                             vdwioffsetptr0+vdwjidx0H,
 615                                             &c6_00,&c12_00);
 616
 617             /* COULOMB ELECTROSTATICS */
 618             velec            = _mm256_mul_ps(qq00,rinv00);
 619             felec            = _mm256_mul_ps(velec,rinvsq00);
 620
 621             /* LENNARD-JONES DISPERSION/REPULSION */
 622
 623             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 624             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 625
 626             fscal            = _mm256_add_ps(felec,fvdw);
 627
 628             /* Calculate temporary vectorial force */
 629             tx               = _mm256_mul_ps(fscal,dx00);
 630             ty               = _mm256_mul_ps(fscal,dy00);
 631             tz               = _mm256_mul_ps(fscal,dz00);
 632
 633             /* Update vectorial force */
 634             fix0             = _mm256_add_ps(fix0,tx);
 635             fiy0             = _mm256_add_ps(fiy0,ty);
 636             fiz0             = _mm256_add_ps(fiz0,tz);
 637
 638             fjptrA             = f+j_coord_offsetA;
 639             fjptrB             = f+j_coord_offsetB;
 640             fjptrC             = f+j_coord_offsetC;
 641             fjptrD             = f+j_coord_offsetD;
 642             fjptrE             = f+j_coord_offsetE;
 643             fjptrF             = f+j_coord_offsetF;
 644             fjptrG             = f+j_coord_offsetG;
 645             fjptrH             = f+j_coord_offsetH;
 646             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 647
 648             /* Inner loop uses 33 flops */
 649         }
 650
 651         if(jidx<j_index_end)
 652         {
 653
 654             /* Get j neighbor index, and coordinate index */
 655             jnrlistA         = jjnr[jidx];
 656             jnrlistB         = jjnr[jidx+1];
 657             jnrlistC         = jjnr[jidx+2];
 658             jnrlistD         = jjnr[jidx+3];
 659             jnrlistE         = jjnr[jidx+4];
 660             jnrlistF         = jjnr[jidx+5];
 661             jnrlistG         = jjnr[jidx+6];
 662             jnrlistH         = jjnr[jidx+7];
 663             /* Sign of each element will be negative for non-real atoms.
 664              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 665              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 666              */
 667             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 668                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 669
 670             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 671             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 672             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 673             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 674             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 675             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 676             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 677             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 678             j_coord_offsetA  = DIM*jnrA;
 679             j_coord_offsetB  = DIM*jnrB;
 680             j_coord_offsetC  = DIM*jnrC;
 681             j_coord_offsetD  = DIM*jnrD;
 682             j_coord_offsetE  = DIM*jnrE;
 683             j_coord_offsetF  = DIM*jnrF;
 684             j_coord_offsetG  = DIM*jnrG;
 685             j_coord_offsetH  = DIM*jnrH;
 686
 687             /* load j atom coordinates */
 688             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 689                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 690                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 691                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 692                                                  &jx0,&jy0,&jz0);
 693
 694             /* Calculate displacement vector */
 695             dx00             = _mm256_sub_ps(ix0,jx0);
 696             dy00             = _mm256_sub_ps(iy0,jy0);
 697             dz00             = _mm256_sub_ps(iz0,jz0);
 698
 699             /* Calculate squared distance and things based on it */
 700             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 701
 702             rinv00           = avx256_invsqrt_f(rsq00);
 703
 704             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 705
 706             /* Load parameters for j particles */
 707             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 708                                                                  charge+jnrC+0,charge+jnrD+0,
 709                                                                  charge+jnrE+0,charge+jnrF+0,
 710                                                                  charge+jnrG+0,charge+jnrH+0);
 711             vdwjidx0A        = 2*vdwtype[jnrA+0];
 712             vdwjidx0B        = 2*vdwtype[jnrB+0];
 713             vdwjidx0C        = 2*vdwtype[jnrC+0];
 714             vdwjidx0D        = 2*vdwtype[jnrD+0];
 715             vdwjidx0E        = 2*vdwtype[jnrE+0];
 716             vdwjidx0F        = 2*vdwtype[jnrF+0];
 717             vdwjidx0G        = 2*vdwtype[jnrG+0];
 718             vdwjidx0H        = 2*vdwtype[jnrH+0];
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             /* Compute parameters for interactions between i and j atoms */
 725             qq00             = _mm256_mul_ps(iq0,jq0);
 726             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 727                                             vdwioffsetptr0+vdwjidx0B,
 728                                             vdwioffsetptr0+vdwjidx0C,
 729                                             vdwioffsetptr0+vdwjidx0D,
 730                                             vdwioffsetptr0+vdwjidx0E,
 731                                             vdwioffsetptr0+vdwjidx0F,
 732                                             vdwioffsetptr0+vdwjidx0G,
 733                                             vdwioffsetptr0+vdwjidx0H,
 734                                             &c6_00,&c12_00);
 735
 736             /* COULOMB ELECTROSTATICS */
 737             velec            = _mm256_mul_ps(qq00,rinv00);
 738             felec            = _mm256_mul_ps(velec,rinvsq00);
 739
 740             /* LENNARD-JONES DISPERSION/REPULSION */
 741
 742             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 743             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 744
 745             fscal            = _mm256_add_ps(felec,fvdw);
 746
 747             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 748
 749             /* Calculate temporary vectorial force */
 750             tx               = _mm256_mul_ps(fscal,dx00);
 751             ty               = _mm256_mul_ps(fscal,dy00);
 752             tz               = _mm256_mul_ps(fscal,dz00);
 753
 754             /* Update vectorial force */
 755             fix0             = _mm256_add_ps(fix0,tx);
 756             fiy0             = _mm256_add_ps(fiy0,ty);
 757             fiz0             = _mm256_add_ps(fiz0,tz);
 758
 759             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 760             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 761             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 762             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 763             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 764             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 765             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 766             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 767             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 768
 769             /* Inner loop uses 33 flops */
 770         }
 771
 772         /* End of innermost loop */
 773
 774         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 775                                                  f+i_coord_offset,fshift+i_shift_offset);
 776
 777         /* Increment number of inner iterations */
 778         inneriter                  += j_index_end - j_index_start;
 779
 780         /* Outer loop uses 7 flops */
 781     }
 782
 783     /* Increment number of outer iterations */
 784     outeriter        += nri;
 785
 786     /* Update outer/inner flops */
 787
 788     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*33);
 789 }