src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwLJ_GeomP1P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  75     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  84     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  85     __m256           dummy_mask,cutoff_mask;
  86     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  87     __m256           one     = _mm256_set1_ps(1.0);
  88     __m256           two     = _mm256_set1_ps(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm256_set1_ps(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102     krf              = _mm256_set1_ps(fr->ic->k_rf);
 103     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 104     crf              = _mm256_set1_ps(fr->ic->c_rf);
 105     nvdwtype         = fr->ntype;
 106     vdwparam         = fr->nbfp;
 107     vdwtype          = mdatoms->typeA;
 108
 109     /* Avoid stupid compiler warnings */
 110     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 111     j_coord_offsetA = 0;
 112     j_coord_offsetB = 0;
 113     j_coord_offsetC = 0;
 114     j_coord_offsetD = 0;
 115     j_coord_offsetE = 0;
 116     j_coord_offsetF = 0;
 117     j_coord_offsetG = 0;
 118     j_coord_offsetH = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     for(iidx=0;iidx<4*DIM;iidx++)
 124     {
 125         scratch[iidx] = 0.0;
 126     }
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm256_setzero_ps();
 146         fiy0             = _mm256_setzero_ps();
 147         fiz0             = _mm256_setzero_ps();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 151         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         velecsum         = _mm256_setzero_ps();
 155         vvdwsum          = _mm256_setzero_ps();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             jnrC             = jjnr[jidx+2];
 165             jnrD             = jjnr[jidx+3];
 166             jnrE             = jjnr[jidx+4];
 167             jnrF             = jjnr[jidx+5];
 168             jnrG             = jjnr[jidx+6];
 169             jnrH             = jjnr[jidx+7];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172             j_coord_offsetC  = DIM*jnrC;
 173             j_coord_offsetD  = DIM*jnrD;
 174             j_coord_offsetE  = DIM*jnrE;
 175             j_coord_offsetF  = DIM*jnrF;
 176             j_coord_offsetG  = DIM*jnrG;
 177             j_coord_offsetH  = DIM*jnrH;
 178
 179             /* load j atom coordinates */
 180             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 182                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 183                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 184                                                  &jx0,&jy0,&jz0);
 185
 186             /* Calculate displacement vector */
 187             dx00             = _mm256_sub_ps(ix0,jx0);
 188             dy00             = _mm256_sub_ps(iy0,jy0);
 189             dz00             = _mm256_sub_ps(iz0,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 193
 194             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 195
 196             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 200                                                                  charge+jnrC+0,charge+jnrD+0,
 201                                                                  charge+jnrE+0,charge+jnrF+0,
 202                                                                  charge+jnrG+0,charge+jnrH+0);
 203             vdwjidx0A        = 2*vdwtype[jnrA+0];
 204             vdwjidx0B        = 2*vdwtype[jnrB+0];
 205             vdwjidx0C        = 2*vdwtype[jnrC+0];
 206             vdwjidx0D        = 2*vdwtype[jnrD+0];
 207             vdwjidx0E        = 2*vdwtype[jnrE+0];
 208             vdwjidx0F        = 2*vdwtype[jnrF+0];
 209             vdwjidx0G        = 2*vdwtype[jnrG+0];
 210             vdwjidx0H        = 2*vdwtype[jnrH+0];
 211
 212             /**************************
 213              * CALCULATE INTERACTIONS *
 214              **************************/
 215
 216             /* Compute parameters for interactions between i and j atoms */
 217             qq00             = _mm256_mul_ps(iq0,jq0);
 218             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 219                                             vdwioffsetptr0+vdwjidx0B,
 220                                             vdwioffsetptr0+vdwjidx0C,
 221                                             vdwioffsetptr0+vdwjidx0D,
 222                                             vdwioffsetptr0+vdwjidx0E,
 223                                             vdwioffsetptr0+vdwjidx0F,
 224                                             vdwioffsetptr0+vdwjidx0G,
 225                                             vdwioffsetptr0+vdwjidx0H,
 226                                             &c6_00,&c12_00);
 227
 228             /* REACTION-FIELD ELECTROSTATICS */
 229             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 230             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 231
 232             /* LENNARD-JONES DISPERSION/REPULSION */
 233
 234             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 235             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 236             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 237             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 238             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 239
 240             /* Update potential sum for this i atom from the interaction with this j atom. */
 241             velecsum         = _mm256_add_ps(velecsum,velec);
 242             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 243
 244             fscal            = _mm256_add_ps(felec,fvdw);
 245
 246             /* Calculate temporary vectorial force */
 247             tx               = _mm256_mul_ps(fscal,dx00);
 248             ty               = _mm256_mul_ps(fscal,dy00);
 249             tz               = _mm256_mul_ps(fscal,dz00);
 250
 251             /* Update vectorial force */
 252             fix0             = _mm256_add_ps(fix0,tx);
 253             fiy0             = _mm256_add_ps(fiy0,ty);
 254             fiz0             = _mm256_add_ps(fiz0,tz);
 255
 256             fjptrA             = f+j_coord_offsetA;
 257             fjptrB             = f+j_coord_offsetB;
 258             fjptrC             = f+j_coord_offsetC;
 259             fjptrD             = f+j_coord_offsetD;
 260             fjptrE             = f+j_coord_offsetE;
 261             fjptrF             = f+j_coord_offsetF;
 262             fjptrG             = f+j_coord_offsetG;
 263             fjptrH             = f+j_coord_offsetH;
 264             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 265
 266             /* Inner loop uses 44 flops */
 267         }
 268
 269         if(jidx<j_index_end)
 270         {
 271
 272             /* Get j neighbor index, and coordinate index */
 273             jnrlistA         = jjnr[jidx];
 274             jnrlistB         = jjnr[jidx+1];
 275             jnrlistC         = jjnr[jidx+2];
 276             jnrlistD         = jjnr[jidx+3];
 277             jnrlistE         = jjnr[jidx+4];
 278             jnrlistF         = jjnr[jidx+5];
 279             jnrlistG         = jjnr[jidx+6];
 280             jnrlistH         = jjnr[jidx+7];
 281             /* Sign of each element will be negative for non-real atoms.
 282              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 283              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 284              */
 285             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 286                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 287
 288             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 289             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 290             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 291             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 292             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 293             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 294             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 295             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 296             j_coord_offsetA  = DIM*jnrA;
 297             j_coord_offsetB  = DIM*jnrB;
 298             j_coord_offsetC  = DIM*jnrC;
 299             j_coord_offsetD  = DIM*jnrD;
 300             j_coord_offsetE  = DIM*jnrE;
 301             j_coord_offsetF  = DIM*jnrF;
 302             j_coord_offsetG  = DIM*jnrG;
 303             j_coord_offsetH  = DIM*jnrH;
 304
 305             /* load j atom coordinates */
 306             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 307                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 308                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 309                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 310                                                  &jx0,&jy0,&jz0);
 311
 312             /* Calculate displacement vector */
 313             dx00             = _mm256_sub_ps(ix0,jx0);
 314             dy00             = _mm256_sub_ps(iy0,jy0);
 315             dz00             = _mm256_sub_ps(iz0,jz0);
 316
 317             /* Calculate squared distance and things based on it */
 318             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 319
 320             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 321
 322             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 323
 324             /* Load parameters for j particles */
 325             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 326                                                                  charge+jnrC+0,charge+jnrD+0,
 327                                                                  charge+jnrE+0,charge+jnrF+0,
 328                                                                  charge+jnrG+0,charge+jnrH+0);
 329             vdwjidx0A        = 2*vdwtype[jnrA+0];
 330             vdwjidx0B        = 2*vdwtype[jnrB+0];
 331             vdwjidx0C        = 2*vdwtype[jnrC+0];
 332             vdwjidx0D        = 2*vdwtype[jnrD+0];
 333             vdwjidx0E        = 2*vdwtype[jnrE+0];
 334             vdwjidx0F        = 2*vdwtype[jnrF+0];
 335             vdwjidx0G        = 2*vdwtype[jnrG+0];
 336             vdwjidx0H        = 2*vdwtype[jnrH+0];
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq00             = _mm256_mul_ps(iq0,jq0);
 344             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 345                                             vdwioffsetptr0+vdwjidx0B,
 346                                             vdwioffsetptr0+vdwjidx0C,
 347                                             vdwioffsetptr0+vdwjidx0D,
 348                                             vdwioffsetptr0+vdwjidx0E,
 349                                             vdwioffsetptr0+vdwjidx0F,
 350                                             vdwioffsetptr0+vdwjidx0G,
 351                                             vdwioffsetptr0+vdwjidx0H,
 352                                             &c6_00,&c12_00);
 353
 354             /* REACTION-FIELD ELECTROSTATICS */
 355             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 356             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 357
 358             /* LENNARD-JONES DISPERSION/REPULSION */
 359
 360             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 361             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 362             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 363             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 364             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 365
 366             /* Update potential sum for this i atom from the interaction with this j atom. */
 367             velec            = _mm256_andnot_ps(dummy_mask,velec);
 368             velecsum         = _mm256_add_ps(velecsum,velec);
 369             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 370             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 371
 372             fscal            = _mm256_add_ps(felec,fvdw);
 373
 374             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm256_mul_ps(fscal,dx00);
 378             ty               = _mm256_mul_ps(fscal,dy00);
 379             tz               = _mm256_mul_ps(fscal,dz00);
 380
 381             /* Update vectorial force */
 382             fix0             = _mm256_add_ps(fix0,tx);
 383             fiy0             = _mm256_add_ps(fiy0,ty);
 384             fiz0             = _mm256_add_ps(fiz0,tz);
 385
 386             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 387             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 388             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 389             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 390             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 391             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 392             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 393             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 394             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 395
 396             /* Inner loop uses 44 flops */
 397         }
 398
 399         /* End of innermost loop */
 400
 401         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 402                                                  f+i_coord_offset,fshift+i_shift_offset);
 403
 404         ggid                        = gid[iidx];
 405         /* Update potential energies */
 406         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 407         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 408
 409         /* Increment number of inner iterations */
 410         inneriter                  += j_index_end - j_index_start;
 411
 412         /* Outer loop uses 9 flops */
 413     }
 414
 415     /* Increment number of outer iterations */
 416     outeriter        += nri;
 417
 418     /* Update outer/inner flops */
 419
 420     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
 421 }
 422 /*
 423  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_single
 424  * Electrostatics interaction: ReactionField
 425  * VdW interaction:            LennardJones
 426  * Geometry:                   Particle-Particle
 427  * Calculate force/pot:        Force
 428  */
 429 void
 430 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_single
 431                     (t_nblist * gmx_restrict                nlist,
 432                      rvec * gmx_restrict                    xx,
 433                      rvec * gmx_restrict                    ff,
 434                      t_forcerec * gmx_restrict              fr,
 435                      t_mdatoms * gmx_restrict               mdatoms,
 436                      nb_kernel_data_t * gmx_restrict        kernel_data,
 437                      t_nrnb * gmx_restrict                  nrnb)
 438 {
 439     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 440      * just 0 for non-waters.
 441      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 442      * jnr indices corresponding to data put in the four positions in the SIMD register.
 443      */
 444     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 445     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 446     int              jnrA,jnrB,jnrC,jnrD;
 447     int              jnrE,jnrF,jnrG,jnrH;
 448     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 449     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 450     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 451     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 452     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 453     real             rcutoff_scalar;
 454     real             *shiftvec,*fshift,*x,*f;
 455     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 456     real             scratch[4*DIM];
 457     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 458     real *           vdwioffsetptr0;
 459     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 460     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 461     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 462     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 463     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 464     real             *charge;
 465     int              nvdwtype;
 466     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 467     int              *vdwtype;
 468     real             *vdwparam;
 469     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 470     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 471     __m256           dummy_mask,cutoff_mask;
 472     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 473     __m256           one     = _mm256_set1_ps(1.0);
 474     __m256           two     = _mm256_set1_ps(2.0);
 475     x                = xx[0];
 476     f                = ff[0];
 477
 478     nri              = nlist->nri;
 479     iinr             = nlist->iinr;
 480     jindex           = nlist->jindex;
 481     jjnr             = nlist->jjnr;
 482     shiftidx         = nlist->shift;
 483     gid              = nlist->gid;
 484     shiftvec         = fr->shift_vec[0];
 485     fshift           = fr->fshift[0];
 486     facel            = _mm256_set1_ps(fr->epsfac);
 487     charge           = mdatoms->chargeA;
 488     krf              = _mm256_set1_ps(fr->ic->k_rf);
 489     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 490     crf              = _mm256_set1_ps(fr->ic->c_rf);
 491     nvdwtype         = fr->ntype;
 492     vdwparam         = fr->nbfp;
 493     vdwtype          = mdatoms->typeA;
 494
 495     /* Avoid stupid compiler warnings */
 496     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 497     j_coord_offsetA = 0;
 498     j_coord_offsetB = 0;
 499     j_coord_offsetC = 0;
 500     j_coord_offsetD = 0;
 501     j_coord_offsetE = 0;
 502     j_coord_offsetF = 0;
 503     j_coord_offsetG = 0;
 504     j_coord_offsetH = 0;
 505
 506     outeriter        = 0;
 507     inneriter        = 0;
 508
 509     for(iidx=0;iidx<4*DIM;iidx++)
 510     {
 511         scratch[iidx] = 0.0;
 512     }
 513
 514     /* Start outer loop over neighborlists */
 515     for(iidx=0; iidx<nri; iidx++)
 516     {
 517         /* Load shift vector for this list */
 518         i_shift_offset   = DIM*shiftidx[iidx];
 519
 520         /* Load limits for loop over neighbors */
 521         j_index_start    = jindex[iidx];
 522         j_index_end      = jindex[iidx+1];
 523
 524         /* Get outer coordinate index */
 525         inr              = iinr[iidx];
 526         i_coord_offset   = DIM*inr;
 527
 528         /* Load i particle coords and add shift vector */
 529         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 530
 531         fix0             = _mm256_setzero_ps();
 532         fiy0             = _mm256_setzero_ps();
 533         fiz0             = _mm256_setzero_ps();
 534
 535         /* Load parameters for i particles */
 536         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 537         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 538
 539         /* Start inner kernel loop */
 540         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 541         {
 542
 543             /* Get j neighbor index, and coordinate index */
 544             jnrA             = jjnr[jidx];
 545             jnrB             = jjnr[jidx+1];
 546             jnrC             = jjnr[jidx+2];
 547             jnrD             = jjnr[jidx+3];
 548             jnrE             = jjnr[jidx+4];
 549             jnrF             = jjnr[jidx+5];
 550             jnrG             = jjnr[jidx+6];
 551             jnrH             = jjnr[jidx+7];
 552             j_coord_offsetA  = DIM*jnrA;
 553             j_coord_offsetB  = DIM*jnrB;
 554             j_coord_offsetC  = DIM*jnrC;
 555             j_coord_offsetD  = DIM*jnrD;
 556             j_coord_offsetE  = DIM*jnrE;
 557             j_coord_offsetF  = DIM*jnrF;
 558             j_coord_offsetG  = DIM*jnrG;
 559             j_coord_offsetH  = DIM*jnrH;
 560
 561             /* load j atom coordinates */
 562             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 563                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 564                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 565                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 566                                                  &jx0,&jy0,&jz0);
 567
 568             /* Calculate displacement vector */
 569             dx00             = _mm256_sub_ps(ix0,jx0);
 570             dy00             = _mm256_sub_ps(iy0,jy0);
 571             dz00             = _mm256_sub_ps(iz0,jz0);
 572
 573             /* Calculate squared distance and things based on it */
 574             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 575
 576             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 577
 578             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 579
 580             /* Load parameters for j particles */
 581             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 582                                                                  charge+jnrC+0,charge+jnrD+0,
 583                                                                  charge+jnrE+0,charge+jnrF+0,
 584                                                                  charge+jnrG+0,charge+jnrH+0);
 585             vdwjidx0A        = 2*vdwtype[jnrA+0];
 586             vdwjidx0B        = 2*vdwtype[jnrB+0];
 587             vdwjidx0C        = 2*vdwtype[jnrC+0];
 588             vdwjidx0D        = 2*vdwtype[jnrD+0];
 589             vdwjidx0E        = 2*vdwtype[jnrE+0];
 590             vdwjidx0F        = 2*vdwtype[jnrF+0];
 591             vdwjidx0G        = 2*vdwtype[jnrG+0];
 592             vdwjidx0H        = 2*vdwtype[jnrH+0];
 593
 594             /**************************
 595              * CALCULATE INTERACTIONS *
 596              **************************/
 597
 598             /* Compute parameters for interactions between i and j atoms */
 599             qq00             = _mm256_mul_ps(iq0,jq0);
 600             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 601                                             vdwioffsetptr0+vdwjidx0B,
 602                                             vdwioffsetptr0+vdwjidx0C,
 603                                             vdwioffsetptr0+vdwjidx0D,
 604                                             vdwioffsetptr0+vdwjidx0E,
 605                                             vdwioffsetptr0+vdwjidx0F,
 606                                             vdwioffsetptr0+vdwjidx0G,
 607                                             vdwioffsetptr0+vdwjidx0H,
 608                                             &c6_00,&c12_00);
 609
 610             /* REACTION-FIELD ELECTROSTATICS */
 611             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 612
 613             /* LENNARD-JONES DISPERSION/REPULSION */
 614
 615             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 616             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 617
 618             fscal            = _mm256_add_ps(felec,fvdw);
 619
 620             /* Calculate temporary vectorial force */
 621             tx               = _mm256_mul_ps(fscal,dx00);
 622             ty               = _mm256_mul_ps(fscal,dy00);
 623             tz               = _mm256_mul_ps(fscal,dz00);
 624
 625             /* Update vectorial force */
 626             fix0             = _mm256_add_ps(fix0,tx);
 627             fiy0             = _mm256_add_ps(fiy0,ty);
 628             fiz0             = _mm256_add_ps(fiz0,tz);
 629
 630             fjptrA             = f+j_coord_offsetA;
 631             fjptrB             = f+j_coord_offsetB;
 632             fjptrC             = f+j_coord_offsetC;
 633             fjptrD             = f+j_coord_offsetD;
 634             fjptrE             = f+j_coord_offsetE;
 635             fjptrF             = f+j_coord_offsetF;
 636             fjptrG             = f+j_coord_offsetG;
 637             fjptrH             = f+j_coord_offsetH;
 638             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 639
 640             /* Inner loop uses 34 flops */
 641         }
 642
 643         if(jidx<j_index_end)
 644         {
 645
 646             /* Get j neighbor index, and coordinate index */
 647             jnrlistA         = jjnr[jidx];
 648             jnrlistB         = jjnr[jidx+1];
 649             jnrlistC         = jjnr[jidx+2];
 650             jnrlistD         = jjnr[jidx+3];
 651             jnrlistE         = jjnr[jidx+4];
 652             jnrlistF         = jjnr[jidx+5];
 653             jnrlistG         = jjnr[jidx+6];
 654             jnrlistH         = jjnr[jidx+7];
 655             /* Sign of each element will be negative for non-real atoms.
 656              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 657              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 658              */
 659             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 660                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 661
 662             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 663             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 664             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 665             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 666             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 667             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 668             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 669             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 670             j_coord_offsetA  = DIM*jnrA;
 671             j_coord_offsetB  = DIM*jnrB;
 672             j_coord_offsetC  = DIM*jnrC;
 673             j_coord_offsetD  = DIM*jnrD;
 674             j_coord_offsetE  = DIM*jnrE;
 675             j_coord_offsetF  = DIM*jnrF;
 676             j_coord_offsetG  = DIM*jnrG;
 677             j_coord_offsetH  = DIM*jnrH;
 678
 679             /* load j atom coordinates */
 680             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 681                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 682                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 683                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 684                                                  &jx0,&jy0,&jz0);
 685
 686             /* Calculate displacement vector */
 687             dx00             = _mm256_sub_ps(ix0,jx0);
 688             dy00             = _mm256_sub_ps(iy0,jy0);
 689             dz00             = _mm256_sub_ps(iz0,jz0);
 690
 691             /* Calculate squared distance and things based on it */
 692             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 693
 694             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 695
 696             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 697
 698             /* Load parameters for j particles */
 699             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 700                                                                  charge+jnrC+0,charge+jnrD+0,
 701                                                                  charge+jnrE+0,charge+jnrF+0,
 702                                                                  charge+jnrG+0,charge+jnrH+0);
 703             vdwjidx0A        = 2*vdwtype[jnrA+0];
 704             vdwjidx0B        = 2*vdwtype[jnrB+0];
 705             vdwjidx0C        = 2*vdwtype[jnrC+0];
 706             vdwjidx0D        = 2*vdwtype[jnrD+0];
 707             vdwjidx0E        = 2*vdwtype[jnrE+0];
 708             vdwjidx0F        = 2*vdwtype[jnrF+0];
 709             vdwjidx0G        = 2*vdwtype[jnrG+0];
 710             vdwjidx0H        = 2*vdwtype[jnrH+0];
 711
 712             /**************************
 713              * CALCULATE INTERACTIONS *
 714              **************************/
 715
 716             /* Compute parameters for interactions between i and j atoms */
 717             qq00             = _mm256_mul_ps(iq0,jq0);
 718             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 719                                             vdwioffsetptr0+vdwjidx0B,
 720                                             vdwioffsetptr0+vdwjidx0C,
 721                                             vdwioffsetptr0+vdwjidx0D,
 722                                             vdwioffsetptr0+vdwjidx0E,
 723                                             vdwioffsetptr0+vdwjidx0F,
 724                                             vdwioffsetptr0+vdwjidx0G,
 725                                             vdwioffsetptr0+vdwjidx0H,
 726                                             &c6_00,&c12_00);
 727
 728             /* REACTION-FIELD ELECTROSTATICS */
 729             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 730
 731             /* LENNARD-JONES DISPERSION/REPULSION */
 732
 733             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 734             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 735
 736             fscal            = _mm256_add_ps(felec,fvdw);
 737
 738             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 739
 740             /* Calculate temporary vectorial force */
 741             tx               = _mm256_mul_ps(fscal,dx00);
 742             ty               = _mm256_mul_ps(fscal,dy00);
 743             tz               = _mm256_mul_ps(fscal,dz00);
 744
 745             /* Update vectorial force */
 746             fix0             = _mm256_add_ps(fix0,tx);
 747             fiy0             = _mm256_add_ps(fiy0,ty);
 748             fiz0             = _mm256_add_ps(fiz0,tz);
 749
 750             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 751             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 752             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 753             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 754             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 755             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 756             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 757             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 758             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 759
 760             /* Inner loop uses 34 flops */
 761         }
 762
 763         /* End of innermost loop */
 764
 765         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 766                                                  f+i_coord_offset,fshift+i_shift_offset);
 767
 768         /* Increment number of inner iterations */
 769         inneriter                  += j_index_end - j_index_start;
 770
 771         /* Outer loop uses 7 flops */
 772     }
 773
 774     /* Increment number of outer iterations */
 775     outeriter        += nri;
 776
 777     /* Update outer/inner flops */
 778
 779     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 780 }