src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwLJ_GeomW4P1_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_ElecCoul_VdwLJ_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_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     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     real *           vdwioffsetptr3;
  79     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  80     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  81     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  82     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  83     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  84     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  85     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  86     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     int              nvdwtype;
  89     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  93     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  94     __m256           dummy_mask,cutoff_mask;
  95     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  96     __m256           one     = _mm256_set1_ps(1.0);
  97     __m256           two     = _mm256_set1_ps(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm256_set1_ps(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     /* Setup water-specific parameters */
 116     inr              = nlist->iinr[0];
 117     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 118     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 119     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 120     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128     j_coord_offsetE = 0;
 129     j_coord_offsetF = 0;
 130     j_coord_offsetG = 0;
 131     j_coord_offsetH = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 158
 159         fix0             = _mm256_setzero_ps();
 160         fiy0             = _mm256_setzero_ps();
 161         fiz0             = _mm256_setzero_ps();
 162         fix1             = _mm256_setzero_ps();
 163         fiy1             = _mm256_setzero_ps();
 164         fiz1             = _mm256_setzero_ps();
 165         fix2             = _mm256_setzero_ps();
 166         fiy2             = _mm256_setzero_ps();
 167         fiz2             = _mm256_setzero_ps();
 168         fix3             = _mm256_setzero_ps();
 169         fiy3             = _mm256_setzero_ps();
 170         fiz3             = _mm256_setzero_ps();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm256_setzero_ps();
 174         vvdwsum          = _mm256_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             jnrE             = jjnr[jidx+4];
 186             jnrF             = jjnr[jidx+5];
 187             jnrG             = jjnr[jidx+6];
 188             jnrH             = jjnr[jidx+7];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193             j_coord_offsetE  = DIM*jnrE;
 194             j_coord_offsetF  = DIM*jnrF;
 195             j_coord_offsetG  = DIM*jnrG;
 196             j_coord_offsetH  = DIM*jnrH;
 197
 198             /* load j atom coordinates */
 199             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 201                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 202                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 203                                                  &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm256_sub_ps(ix0,jx0);
 207             dy00             = _mm256_sub_ps(iy0,jy0);
 208             dz00             = _mm256_sub_ps(iz0,jz0);
 209             dx10             = _mm256_sub_ps(ix1,jx0);
 210             dy10             = _mm256_sub_ps(iy1,jy0);
 211             dz10             = _mm256_sub_ps(iz1,jz0);
 212             dx20             = _mm256_sub_ps(ix2,jx0);
 213             dy20             = _mm256_sub_ps(iy2,jy0);
 214             dz20             = _mm256_sub_ps(iz2,jz0);
 215             dx30             = _mm256_sub_ps(ix3,jx0);
 216             dy30             = _mm256_sub_ps(iy3,jy0);
 217             dz30             = _mm256_sub_ps(iz3,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 221             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 222             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 223             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 224
 225             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 226             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 227             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 228
 229             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 230             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 231             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 232             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 236                                                                  charge+jnrC+0,charge+jnrD+0,
 237                                                                  charge+jnrE+0,charge+jnrF+0,
 238                                                                  charge+jnrG+0,charge+jnrH+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241             vdwjidx0C        = 2*vdwtype[jnrC+0];
 242             vdwjidx0D        = 2*vdwtype[jnrD+0];
 243             vdwjidx0E        = 2*vdwtype[jnrE+0];
 244             vdwjidx0F        = 2*vdwtype[jnrF+0];
 245             vdwjidx0G        = 2*vdwtype[jnrG+0];
 246             vdwjidx0H        = 2*vdwtype[jnrH+0];
 247
 248             fjx0             = _mm256_setzero_ps();
 249             fjy0             = _mm256_setzero_ps();
 250             fjz0             = _mm256_setzero_ps();
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 258                                             vdwioffsetptr0+vdwjidx0B,
 259                                             vdwioffsetptr0+vdwjidx0C,
 260                                             vdwioffsetptr0+vdwjidx0D,
 261                                             vdwioffsetptr0+vdwjidx0E,
 262                                             vdwioffsetptr0+vdwjidx0F,
 263                                             vdwioffsetptr0+vdwjidx0G,
 264                                             vdwioffsetptr0+vdwjidx0H,
 265                                             &c6_00,&c12_00);
 266
 267             /* LENNARD-JONES DISPERSION/REPULSION */
 268
 269             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 270             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 271             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 272             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 273             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 277
 278             fscal            = fvdw;
 279
 280             /* Calculate temporary vectorial force */
 281             tx               = _mm256_mul_ps(fscal,dx00);
 282             ty               = _mm256_mul_ps(fscal,dy00);
 283             tz               = _mm256_mul_ps(fscal,dz00);
 284
 285             /* Update vectorial force */
 286             fix0             = _mm256_add_ps(fix0,tx);
 287             fiy0             = _mm256_add_ps(fiy0,ty);
 288             fiz0             = _mm256_add_ps(fiz0,tz);
 289
 290             fjx0             = _mm256_add_ps(fjx0,tx);
 291             fjy0             = _mm256_add_ps(fjy0,ty);
 292             fjz0             = _mm256_add_ps(fjz0,tz);
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq10             = _mm256_mul_ps(iq1,jq0);
 300
 301             /* COULOMB ELECTROSTATICS */
 302             velec            = _mm256_mul_ps(qq10,rinv10);
 303             felec            = _mm256_mul_ps(velec,rinvsq10);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             velecsum         = _mm256_add_ps(velecsum,velec);
 307
 308             fscal            = felec;
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm256_mul_ps(fscal,dx10);
 312             ty               = _mm256_mul_ps(fscal,dy10);
 313             tz               = _mm256_mul_ps(fscal,dz10);
 314
 315             /* Update vectorial force */
 316             fix1             = _mm256_add_ps(fix1,tx);
 317             fiy1             = _mm256_add_ps(fiy1,ty);
 318             fiz1             = _mm256_add_ps(fiz1,tz);
 319
 320             fjx0             = _mm256_add_ps(fjx0,tx);
 321             fjy0             = _mm256_add_ps(fjy0,ty);
 322             fjz0             = _mm256_add_ps(fjz0,tz);
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq20             = _mm256_mul_ps(iq2,jq0);
 330
 331             /* COULOMB ELECTROSTATICS */
 332             velec            = _mm256_mul_ps(qq20,rinv20);
 333             felec            = _mm256_mul_ps(velec,rinvsq20);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velecsum         = _mm256_add_ps(velecsum,velec);
 337
 338             fscal            = felec;
 339
 340             /* Calculate temporary vectorial force */
 341             tx               = _mm256_mul_ps(fscal,dx20);
 342             ty               = _mm256_mul_ps(fscal,dy20);
 343             tz               = _mm256_mul_ps(fscal,dz20);
 344
 345             /* Update vectorial force */
 346             fix2             = _mm256_add_ps(fix2,tx);
 347             fiy2             = _mm256_add_ps(fiy2,ty);
 348             fiz2             = _mm256_add_ps(fiz2,tz);
 349
 350             fjx0             = _mm256_add_ps(fjx0,tx);
 351             fjy0             = _mm256_add_ps(fjy0,ty);
 352             fjz0             = _mm256_add_ps(fjz0,tz);
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq30             = _mm256_mul_ps(iq3,jq0);
 360
 361             /* COULOMB ELECTROSTATICS */
 362             velec            = _mm256_mul_ps(qq30,rinv30);
 363             felec            = _mm256_mul_ps(velec,rinvsq30);
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velecsum         = _mm256_add_ps(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             /* Calculate temporary vectorial force */
 371             tx               = _mm256_mul_ps(fscal,dx30);
 372             ty               = _mm256_mul_ps(fscal,dy30);
 373             tz               = _mm256_mul_ps(fscal,dz30);
 374
 375             /* Update vectorial force */
 376             fix3             = _mm256_add_ps(fix3,tx);
 377             fiy3             = _mm256_add_ps(fiy3,ty);
 378             fiz3             = _mm256_add_ps(fiz3,tz);
 379
 380             fjx0             = _mm256_add_ps(fjx0,tx);
 381             fjy0             = _mm256_add_ps(fjy0,ty);
 382             fjz0             = _mm256_add_ps(fjz0,tz);
 383
 384             fjptrA             = f+j_coord_offsetA;
 385             fjptrB             = f+j_coord_offsetB;
 386             fjptrC             = f+j_coord_offsetC;
 387             fjptrD             = f+j_coord_offsetD;
 388             fjptrE             = f+j_coord_offsetE;
 389             fjptrF             = f+j_coord_offsetF;
 390             fjptrG             = f+j_coord_offsetG;
 391             fjptrH             = f+j_coord_offsetH;
 392
 393             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 394
 395             /* Inner loop uses 116 flops */
 396         }
 397
 398         if(jidx<j_index_end)
 399         {
 400
 401             /* Get j neighbor index, and coordinate index */
 402             jnrlistA         = jjnr[jidx];
 403             jnrlistB         = jjnr[jidx+1];
 404             jnrlistC         = jjnr[jidx+2];
 405             jnrlistD         = jjnr[jidx+3];
 406             jnrlistE         = jjnr[jidx+4];
 407             jnrlistF         = jjnr[jidx+5];
 408             jnrlistG         = jjnr[jidx+6];
 409             jnrlistH         = jjnr[jidx+7];
 410             /* Sign of each element will be negative for non-real atoms.
 411              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 412              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 413              */
 414             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 415                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 416
 417             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 418             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 419             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 420             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 421             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 422             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 423             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 424             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 425             j_coord_offsetA  = DIM*jnrA;
 426             j_coord_offsetB  = DIM*jnrB;
 427             j_coord_offsetC  = DIM*jnrC;
 428             j_coord_offsetD  = DIM*jnrD;
 429             j_coord_offsetE  = DIM*jnrE;
 430             j_coord_offsetF  = DIM*jnrF;
 431             j_coord_offsetG  = DIM*jnrG;
 432             j_coord_offsetH  = DIM*jnrH;
 433
 434             /* load j atom coordinates */
 435             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 436                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 437                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 438                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 439                                                  &jx0,&jy0,&jz0);
 440
 441             /* Calculate displacement vector */
 442             dx00             = _mm256_sub_ps(ix0,jx0);
 443             dy00             = _mm256_sub_ps(iy0,jy0);
 444             dz00             = _mm256_sub_ps(iz0,jz0);
 445             dx10             = _mm256_sub_ps(ix1,jx0);
 446             dy10             = _mm256_sub_ps(iy1,jy0);
 447             dz10             = _mm256_sub_ps(iz1,jz0);
 448             dx20             = _mm256_sub_ps(ix2,jx0);
 449             dy20             = _mm256_sub_ps(iy2,jy0);
 450             dz20             = _mm256_sub_ps(iz2,jz0);
 451             dx30             = _mm256_sub_ps(ix3,jx0);
 452             dy30             = _mm256_sub_ps(iy3,jy0);
 453             dz30             = _mm256_sub_ps(iz3,jz0);
 454
 455             /* Calculate squared distance and things based on it */
 456             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 457             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 458             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 459             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 460
 461             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 462             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 463             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 464
 465             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 466             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 467             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 468             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 469
 470             /* Load parameters for j particles */
 471             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 472                                                                  charge+jnrC+0,charge+jnrD+0,
 473                                                                  charge+jnrE+0,charge+jnrF+0,
 474                                                                  charge+jnrG+0,charge+jnrH+0);
 475             vdwjidx0A        = 2*vdwtype[jnrA+0];
 476             vdwjidx0B        = 2*vdwtype[jnrB+0];
 477             vdwjidx0C        = 2*vdwtype[jnrC+0];
 478             vdwjidx0D        = 2*vdwtype[jnrD+0];
 479             vdwjidx0E        = 2*vdwtype[jnrE+0];
 480             vdwjidx0F        = 2*vdwtype[jnrF+0];
 481             vdwjidx0G        = 2*vdwtype[jnrG+0];
 482             vdwjidx0H        = 2*vdwtype[jnrH+0];
 483
 484             fjx0             = _mm256_setzero_ps();
 485             fjy0             = _mm256_setzero_ps();
 486             fjz0             = _mm256_setzero_ps();
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             /* Compute parameters for interactions between i and j atoms */
 493             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 494                                             vdwioffsetptr0+vdwjidx0B,
 495                                             vdwioffsetptr0+vdwjidx0C,
 496                                             vdwioffsetptr0+vdwjidx0D,
 497                                             vdwioffsetptr0+vdwjidx0E,
 498                                             vdwioffsetptr0+vdwjidx0F,
 499                                             vdwioffsetptr0+vdwjidx0G,
 500                                             vdwioffsetptr0+vdwjidx0H,
 501                                             &c6_00,&c12_00);
 502
 503             /* LENNARD-JONES DISPERSION/REPULSION */
 504
 505             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 506             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 507             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 508             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 509             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 510
 511             /* Update potential sum for this i atom from the interaction with this j atom. */
 512             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 513             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 514
 515             fscal            = fvdw;
 516
 517             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 518
 519             /* Calculate temporary vectorial force */
 520             tx               = _mm256_mul_ps(fscal,dx00);
 521             ty               = _mm256_mul_ps(fscal,dy00);
 522             tz               = _mm256_mul_ps(fscal,dz00);
 523
 524             /* Update vectorial force */
 525             fix0             = _mm256_add_ps(fix0,tx);
 526             fiy0             = _mm256_add_ps(fiy0,ty);
 527             fiz0             = _mm256_add_ps(fiz0,tz);
 528
 529             fjx0             = _mm256_add_ps(fjx0,tx);
 530             fjy0             = _mm256_add_ps(fjy0,ty);
 531             fjz0             = _mm256_add_ps(fjz0,tz);
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq10             = _mm256_mul_ps(iq1,jq0);
 539
 540             /* COULOMB ELECTROSTATICS */
 541             velec            = _mm256_mul_ps(qq10,rinv10);
 542             felec            = _mm256_mul_ps(velec,rinvsq10);
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velec            = _mm256_andnot_ps(dummy_mask,velec);
 546             velecsum         = _mm256_add_ps(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm256_mul_ps(fscal,dx10);
 554             ty               = _mm256_mul_ps(fscal,dy10);
 555             tz               = _mm256_mul_ps(fscal,dz10);
 556
 557             /* Update vectorial force */
 558             fix1             = _mm256_add_ps(fix1,tx);
 559             fiy1             = _mm256_add_ps(fiy1,ty);
 560             fiz1             = _mm256_add_ps(fiz1,tz);
 561
 562             fjx0             = _mm256_add_ps(fjx0,tx);
 563             fjy0             = _mm256_add_ps(fjy0,ty);
 564             fjz0             = _mm256_add_ps(fjz0,tz);
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             /* Compute parameters for interactions between i and j atoms */
 571             qq20             = _mm256_mul_ps(iq2,jq0);
 572
 573             /* COULOMB ELECTROSTATICS */
 574             velec            = _mm256_mul_ps(qq20,rinv20);
 575             felec            = _mm256_mul_ps(velec,rinvsq20);
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm256_andnot_ps(dummy_mask,velec);
 579             velecsum         = _mm256_add_ps(velecsum,velec);
 580
 581             fscal            = felec;
 582
 583             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm256_mul_ps(fscal,dx20);
 587             ty               = _mm256_mul_ps(fscal,dy20);
 588             tz               = _mm256_mul_ps(fscal,dz20);
 589
 590             /* Update vectorial force */
 591             fix2             = _mm256_add_ps(fix2,tx);
 592             fiy2             = _mm256_add_ps(fiy2,ty);
 593             fiz2             = _mm256_add_ps(fiz2,tz);
 594
 595             fjx0             = _mm256_add_ps(fjx0,tx);
 596             fjy0             = _mm256_add_ps(fjy0,ty);
 597             fjz0             = _mm256_add_ps(fjz0,tz);
 598
 599             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             /* Compute parameters for interactions between i and j atoms */
 604             qq30             = _mm256_mul_ps(iq3,jq0);
 605
 606             /* COULOMB ELECTROSTATICS */
 607             velec            = _mm256_mul_ps(qq30,rinv30);
 608             felec            = _mm256_mul_ps(velec,rinvsq30);
 609
 610             /* Update potential sum for this i atom from the interaction with this j atom. */
 611             velec            = _mm256_andnot_ps(dummy_mask,velec);
 612             velecsum         = _mm256_add_ps(velecsum,velec);
 613
 614             fscal            = felec;
 615
 616             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 617
 618             /* Calculate temporary vectorial force */
 619             tx               = _mm256_mul_ps(fscal,dx30);
 620             ty               = _mm256_mul_ps(fscal,dy30);
 621             tz               = _mm256_mul_ps(fscal,dz30);
 622
 623             /* Update vectorial force */
 624             fix3             = _mm256_add_ps(fix3,tx);
 625             fiy3             = _mm256_add_ps(fiy3,ty);
 626             fiz3             = _mm256_add_ps(fiz3,tz);
 627
 628             fjx0             = _mm256_add_ps(fjx0,tx);
 629             fjy0             = _mm256_add_ps(fjy0,ty);
 630             fjz0             = _mm256_add_ps(fjz0,tz);
 631
 632             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 633             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 634             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 635             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 636             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 637             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 638             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 639             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 640
 641             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 642
 643             /* Inner loop uses 116 flops */
 644         }
 645
 646         /* End of innermost loop */
 647
 648         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 649                                                  f+i_coord_offset,fshift+i_shift_offset);
 650
 651         ggid                        = gid[iidx];
 652         /* Update potential energies */
 653         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 654         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 655
 656         /* Increment number of inner iterations */
 657         inneriter                  += j_index_end - j_index_start;
 658
 659         /* Outer loop uses 26 flops */
 660     }
 661
 662     /* Increment number of outer iterations */
 663     outeriter        += nri;
 664
 665     /* Update outer/inner flops */
 666
 667     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*116);
 668 }
 669 /*
 670  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_single
 671  * Electrostatics interaction: Coulomb
 672  * VdW interaction:            LennardJones
 673  * Geometry:                   Water4-Particle
 674  * Calculate force/pot:        Force
 675  */
 676 void
 677 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_256_single
 678                     (t_nblist * gmx_restrict                nlist,
 679                      rvec * gmx_restrict                    xx,
 680                      rvec * gmx_restrict                    ff,
 681                      t_forcerec * gmx_restrict              fr,
 682                      t_mdatoms * gmx_restrict               mdatoms,
 683                      nb_kernel_data_t * gmx_restrict        kernel_data,
 684                      t_nrnb * gmx_restrict                  nrnb)
 685 {
 686     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 687      * just 0 for non-waters.
 688      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 689      * jnr indices corresponding to data put in the four positions in the SIMD register.
 690      */
 691     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 692     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 693     int              jnrA,jnrB,jnrC,jnrD;
 694     int              jnrE,jnrF,jnrG,jnrH;
 695     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 696     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 697     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 698     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 699     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 700     real             rcutoff_scalar;
 701     real             *shiftvec,*fshift,*x,*f;
 702     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 703     real             scratch[4*DIM];
 704     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 705     real *           vdwioffsetptr0;
 706     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 707     real *           vdwioffsetptr1;
 708     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 709     real *           vdwioffsetptr2;
 710     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 711     real *           vdwioffsetptr3;
 712     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 713     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 714     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 715     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 716     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 717     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 718     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 719     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 720     real             *charge;
 721     int              nvdwtype;
 722     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 723     int              *vdwtype;
 724     real             *vdwparam;
 725     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 726     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 727     __m256           dummy_mask,cutoff_mask;
 728     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 729     __m256           one     = _mm256_set1_ps(1.0);
 730     __m256           two     = _mm256_set1_ps(2.0);
 731     x                = xx[0];
 732     f                = ff[0];
 733
 734     nri              = nlist->nri;
 735     iinr             = nlist->iinr;
 736     jindex           = nlist->jindex;
 737     jjnr             = nlist->jjnr;
 738     shiftidx         = nlist->shift;
 739     gid              = nlist->gid;
 740     shiftvec         = fr->shift_vec[0];
 741     fshift           = fr->fshift[0];
 742     facel            = _mm256_set1_ps(fr->epsfac);
 743     charge           = mdatoms->chargeA;
 744     nvdwtype         = fr->ntype;
 745     vdwparam         = fr->nbfp;
 746     vdwtype          = mdatoms->typeA;
 747
 748     /* Setup water-specific parameters */
 749     inr              = nlist->iinr[0];
 750     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 751     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 752     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 753     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 754
 755     /* Avoid stupid compiler warnings */
 756     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 757     j_coord_offsetA = 0;
 758     j_coord_offsetB = 0;
 759     j_coord_offsetC = 0;
 760     j_coord_offsetD = 0;
 761     j_coord_offsetE = 0;
 762     j_coord_offsetF = 0;
 763     j_coord_offsetG = 0;
 764     j_coord_offsetH = 0;
 765
 766     outeriter        = 0;
 767     inneriter        = 0;
 768
 769     for(iidx=0;iidx<4*DIM;iidx++)
 770     {
 771         scratch[iidx] = 0.0;
 772     }
 773
 774     /* Start outer loop over neighborlists */
 775     for(iidx=0; iidx<nri; iidx++)
 776     {
 777         /* Load shift vector for this list */
 778         i_shift_offset   = DIM*shiftidx[iidx];
 779
 780         /* Load limits for loop over neighbors */
 781         j_index_start    = jindex[iidx];
 782         j_index_end      = jindex[iidx+1];
 783
 784         /* Get outer coordinate index */
 785         inr              = iinr[iidx];
 786         i_coord_offset   = DIM*inr;
 787
 788         /* Load i particle coords and add shift vector */
 789         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 790                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 791
 792         fix0             = _mm256_setzero_ps();
 793         fiy0             = _mm256_setzero_ps();
 794         fiz0             = _mm256_setzero_ps();
 795         fix1             = _mm256_setzero_ps();
 796         fiy1             = _mm256_setzero_ps();
 797         fiz1             = _mm256_setzero_ps();
 798         fix2             = _mm256_setzero_ps();
 799         fiy2             = _mm256_setzero_ps();
 800         fiz2             = _mm256_setzero_ps();
 801         fix3             = _mm256_setzero_ps();
 802         fiy3             = _mm256_setzero_ps();
 803         fiz3             = _mm256_setzero_ps();
 804
 805         /* Start inner kernel loop */
 806         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 807         {
 808
 809             /* Get j neighbor index, and coordinate index */
 810             jnrA             = jjnr[jidx];
 811             jnrB             = jjnr[jidx+1];
 812             jnrC             = jjnr[jidx+2];
 813             jnrD             = jjnr[jidx+3];
 814             jnrE             = jjnr[jidx+4];
 815             jnrF             = jjnr[jidx+5];
 816             jnrG             = jjnr[jidx+6];
 817             jnrH             = jjnr[jidx+7];
 818             j_coord_offsetA  = DIM*jnrA;
 819             j_coord_offsetB  = DIM*jnrB;
 820             j_coord_offsetC  = DIM*jnrC;
 821             j_coord_offsetD  = DIM*jnrD;
 822             j_coord_offsetE  = DIM*jnrE;
 823             j_coord_offsetF  = DIM*jnrF;
 824             j_coord_offsetG  = DIM*jnrG;
 825             j_coord_offsetH  = DIM*jnrH;
 826
 827             /* load j atom coordinates */
 828             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 829                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 830                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 831                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 832                                                  &jx0,&jy0,&jz0);
 833
 834             /* Calculate displacement vector */
 835             dx00             = _mm256_sub_ps(ix0,jx0);
 836             dy00             = _mm256_sub_ps(iy0,jy0);
 837             dz00             = _mm256_sub_ps(iz0,jz0);
 838             dx10             = _mm256_sub_ps(ix1,jx0);
 839             dy10             = _mm256_sub_ps(iy1,jy0);
 840             dz10             = _mm256_sub_ps(iz1,jz0);
 841             dx20             = _mm256_sub_ps(ix2,jx0);
 842             dy20             = _mm256_sub_ps(iy2,jy0);
 843             dz20             = _mm256_sub_ps(iz2,jz0);
 844             dx30             = _mm256_sub_ps(ix3,jx0);
 845             dy30             = _mm256_sub_ps(iy3,jy0);
 846             dz30             = _mm256_sub_ps(iz3,jz0);
 847
 848             /* Calculate squared distance and things based on it */
 849             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 850             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 851             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 852             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 853
 854             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 855             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 856             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 857
 858             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 859             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 860             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 861             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 862
 863             /* Load parameters for j particles */
 864             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 865                                                                  charge+jnrC+0,charge+jnrD+0,
 866                                                                  charge+jnrE+0,charge+jnrF+0,
 867                                                                  charge+jnrG+0,charge+jnrH+0);
 868             vdwjidx0A        = 2*vdwtype[jnrA+0];
 869             vdwjidx0B        = 2*vdwtype[jnrB+0];
 870             vdwjidx0C        = 2*vdwtype[jnrC+0];
 871             vdwjidx0D        = 2*vdwtype[jnrD+0];
 872             vdwjidx0E        = 2*vdwtype[jnrE+0];
 873             vdwjidx0F        = 2*vdwtype[jnrF+0];
 874             vdwjidx0G        = 2*vdwtype[jnrG+0];
 875             vdwjidx0H        = 2*vdwtype[jnrH+0];
 876
 877             fjx0             = _mm256_setzero_ps();
 878             fjy0             = _mm256_setzero_ps();
 879             fjz0             = _mm256_setzero_ps();
 880
 881             /**************************
 882              * CALCULATE INTERACTIONS *
 883              **************************/
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 887                                             vdwioffsetptr0+vdwjidx0B,
 888                                             vdwioffsetptr0+vdwjidx0C,
 889                                             vdwioffsetptr0+vdwjidx0D,
 890                                             vdwioffsetptr0+vdwjidx0E,
 891                                             vdwioffsetptr0+vdwjidx0F,
 892                                             vdwioffsetptr0+vdwjidx0G,
 893                                             vdwioffsetptr0+vdwjidx0H,
 894                                             &c6_00,&c12_00);
 895
 896             /* LENNARD-JONES DISPERSION/REPULSION */
 897
 898             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 899             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 900
 901             fscal            = fvdw;
 902
 903             /* Calculate temporary vectorial force */
 904             tx               = _mm256_mul_ps(fscal,dx00);
 905             ty               = _mm256_mul_ps(fscal,dy00);
 906             tz               = _mm256_mul_ps(fscal,dz00);
 907
 908             /* Update vectorial force */
 909             fix0             = _mm256_add_ps(fix0,tx);
 910             fiy0             = _mm256_add_ps(fiy0,ty);
 911             fiz0             = _mm256_add_ps(fiz0,tz);
 912
 913             fjx0             = _mm256_add_ps(fjx0,tx);
 914             fjy0             = _mm256_add_ps(fjy0,ty);
 915             fjz0             = _mm256_add_ps(fjz0,tz);
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             /* Compute parameters for interactions between i and j atoms */
 922             qq10             = _mm256_mul_ps(iq1,jq0);
 923
 924             /* COULOMB ELECTROSTATICS */
 925             velec            = _mm256_mul_ps(qq10,rinv10);
 926             felec            = _mm256_mul_ps(velec,rinvsq10);
 927
 928             fscal            = felec;
 929
 930             /* Calculate temporary vectorial force */
 931             tx               = _mm256_mul_ps(fscal,dx10);
 932             ty               = _mm256_mul_ps(fscal,dy10);
 933             tz               = _mm256_mul_ps(fscal,dz10);
 934
 935             /* Update vectorial force */
 936             fix1             = _mm256_add_ps(fix1,tx);
 937             fiy1             = _mm256_add_ps(fiy1,ty);
 938             fiz1             = _mm256_add_ps(fiz1,tz);
 939
 940             fjx0             = _mm256_add_ps(fjx0,tx);
 941             fjy0             = _mm256_add_ps(fjy0,ty);
 942             fjz0             = _mm256_add_ps(fjz0,tz);
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             /* Compute parameters for interactions between i and j atoms */
 949             qq20             = _mm256_mul_ps(iq2,jq0);
 950
 951             /* COULOMB ELECTROSTATICS */
 952             velec            = _mm256_mul_ps(qq20,rinv20);
 953             felec            = _mm256_mul_ps(velec,rinvsq20);
 954
 955             fscal            = felec;
 956
 957             /* Calculate temporary vectorial force */
 958             tx               = _mm256_mul_ps(fscal,dx20);
 959             ty               = _mm256_mul_ps(fscal,dy20);
 960             tz               = _mm256_mul_ps(fscal,dz20);
 961
 962             /* Update vectorial force */
 963             fix2             = _mm256_add_ps(fix2,tx);
 964             fiy2             = _mm256_add_ps(fiy2,ty);
 965             fiz2             = _mm256_add_ps(fiz2,tz);
 966
 967             fjx0             = _mm256_add_ps(fjx0,tx);
 968             fjy0             = _mm256_add_ps(fjy0,ty);
 969             fjz0             = _mm256_add_ps(fjz0,tz);
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq30             = _mm256_mul_ps(iq3,jq0);
 977
 978             /* COULOMB ELECTROSTATICS */
 979             velec            = _mm256_mul_ps(qq30,rinv30);
 980             felec            = _mm256_mul_ps(velec,rinvsq30);
 981
 982             fscal            = felec;
 983
 984             /* Calculate temporary vectorial force */
 985             tx               = _mm256_mul_ps(fscal,dx30);
 986             ty               = _mm256_mul_ps(fscal,dy30);
 987             tz               = _mm256_mul_ps(fscal,dz30);
 988
 989             /* Update vectorial force */
 990             fix3             = _mm256_add_ps(fix3,tx);
 991             fiy3             = _mm256_add_ps(fiy3,ty);
 992             fiz3             = _mm256_add_ps(fiz3,tz);
 993
 994             fjx0             = _mm256_add_ps(fjx0,tx);
 995             fjy0             = _mm256_add_ps(fjy0,ty);
 996             fjz0             = _mm256_add_ps(fjz0,tz);
 997
 998             fjptrA             = f+j_coord_offsetA;
 999             fjptrB             = f+j_coord_offsetB;
1000             fjptrC             = f+j_coord_offsetC;
1001             fjptrD             = f+j_coord_offsetD;
1002             fjptrE             = f+j_coord_offsetE;
1003             fjptrF             = f+j_coord_offsetF;
1004             fjptrG             = f+j_coord_offsetG;
1005             fjptrH             = f+j_coord_offsetH;
1006
1007             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1008
1009             /* Inner loop uses 108 flops */
1010         }
1011
1012         if(jidx<j_index_end)
1013         {
1014
1015             /* Get j neighbor index, and coordinate index */
1016             jnrlistA         = jjnr[jidx];
1017             jnrlistB         = jjnr[jidx+1];
1018             jnrlistC         = jjnr[jidx+2];
1019             jnrlistD         = jjnr[jidx+3];
1020             jnrlistE         = jjnr[jidx+4];
1021             jnrlistF         = jjnr[jidx+5];
1022             jnrlistG         = jjnr[jidx+6];
1023             jnrlistH         = jjnr[jidx+7];
1024             /* Sign of each element will be negative for non-real atoms.
1025              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1026              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1027              */
1028             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1029                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1030
1031             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1032             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1033             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1034             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1035             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1036             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1037             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1038             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1039             j_coord_offsetA  = DIM*jnrA;
1040             j_coord_offsetB  = DIM*jnrB;
1041             j_coord_offsetC  = DIM*jnrC;
1042             j_coord_offsetD  = DIM*jnrD;
1043             j_coord_offsetE  = DIM*jnrE;
1044             j_coord_offsetF  = DIM*jnrF;
1045             j_coord_offsetG  = DIM*jnrG;
1046             j_coord_offsetH  = DIM*jnrH;
1047
1048             /* load j atom coordinates */
1049             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1050                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1051                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1052                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1053                                                  &jx0,&jy0,&jz0);
1054
1055             /* Calculate displacement vector */
1056             dx00             = _mm256_sub_ps(ix0,jx0);
1057             dy00             = _mm256_sub_ps(iy0,jy0);
1058             dz00             = _mm256_sub_ps(iz0,jz0);
1059             dx10             = _mm256_sub_ps(ix1,jx0);
1060             dy10             = _mm256_sub_ps(iy1,jy0);
1061             dz10             = _mm256_sub_ps(iz1,jz0);
1062             dx20             = _mm256_sub_ps(ix2,jx0);
1063             dy20             = _mm256_sub_ps(iy2,jy0);
1064             dz20             = _mm256_sub_ps(iz2,jz0);
1065             dx30             = _mm256_sub_ps(ix3,jx0);
1066             dy30             = _mm256_sub_ps(iy3,jy0);
1067             dz30             = _mm256_sub_ps(iz3,jz0);
1068
1069             /* Calculate squared distance and things based on it */
1070             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1071             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1072             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1073             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1074
1075             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1076             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1077             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1078
1079             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1080             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1081             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1082             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1083
1084             /* Load parameters for j particles */
1085             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1086                                                                  charge+jnrC+0,charge+jnrD+0,
1087                                                                  charge+jnrE+0,charge+jnrF+0,
1088                                                                  charge+jnrG+0,charge+jnrH+0);
1089             vdwjidx0A        = 2*vdwtype[jnrA+0];
1090             vdwjidx0B        = 2*vdwtype[jnrB+0];
1091             vdwjidx0C        = 2*vdwtype[jnrC+0];
1092             vdwjidx0D        = 2*vdwtype[jnrD+0];
1093             vdwjidx0E        = 2*vdwtype[jnrE+0];
1094             vdwjidx0F        = 2*vdwtype[jnrF+0];
1095             vdwjidx0G        = 2*vdwtype[jnrG+0];
1096             vdwjidx0H        = 2*vdwtype[jnrH+0];
1097
1098             fjx0             = _mm256_setzero_ps();
1099             fjy0             = _mm256_setzero_ps();
1100             fjz0             = _mm256_setzero_ps();
1101
1102             /**************************
1103              * CALCULATE INTERACTIONS *
1104              **************************/
1105
1106             /* Compute parameters for interactions between i and j atoms */
1107             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1108                                             vdwioffsetptr0+vdwjidx0B,
1109                                             vdwioffsetptr0+vdwjidx0C,
1110                                             vdwioffsetptr0+vdwjidx0D,
1111                                             vdwioffsetptr0+vdwjidx0E,
1112                                             vdwioffsetptr0+vdwjidx0F,
1113                                             vdwioffsetptr0+vdwjidx0G,
1114                                             vdwioffsetptr0+vdwjidx0H,
1115                                             &c6_00,&c12_00);
1116
1117             /* LENNARD-JONES DISPERSION/REPULSION */
1118
1119             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1120             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1121
1122             fscal            = fvdw;
1123
1124             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1125
1126             /* Calculate temporary vectorial force */
1127             tx               = _mm256_mul_ps(fscal,dx00);
1128             ty               = _mm256_mul_ps(fscal,dy00);
1129             tz               = _mm256_mul_ps(fscal,dz00);
1130
1131             /* Update vectorial force */
1132             fix0             = _mm256_add_ps(fix0,tx);
1133             fiy0             = _mm256_add_ps(fiy0,ty);
1134             fiz0             = _mm256_add_ps(fiz0,tz);
1135
1136             fjx0             = _mm256_add_ps(fjx0,tx);
1137             fjy0             = _mm256_add_ps(fjy0,ty);
1138             fjz0             = _mm256_add_ps(fjz0,tz);
1139
1140             /**************************
1141              * CALCULATE INTERACTIONS *
1142              **************************/
1143
1144             /* Compute parameters for interactions between i and j atoms */
1145             qq10             = _mm256_mul_ps(iq1,jq0);
1146
1147             /* COULOMB ELECTROSTATICS */
1148             velec            = _mm256_mul_ps(qq10,rinv10);
1149             felec            = _mm256_mul_ps(velec,rinvsq10);
1150
1151             fscal            = felec;
1152
1153             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1154
1155             /* Calculate temporary vectorial force */
1156             tx               = _mm256_mul_ps(fscal,dx10);
1157             ty               = _mm256_mul_ps(fscal,dy10);
1158             tz               = _mm256_mul_ps(fscal,dz10);
1159
1160             /* Update vectorial force */
1161             fix1             = _mm256_add_ps(fix1,tx);
1162             fiy1             = _mm256_add_ps(fiy1,ty);
1163             fiz1             = _mm256_add_ps(fiz1,tz);
1164
1165             fjx0             = _mm256_add_ps(fjx0,tx);
1166             fjy0             = _mm256_add_ps(fjy0,ty);
1167             fjz0             = _mm256_add_ps(fjz0,tz);
1168
1169             /**************************
1170              * CALCULATE INTERACTIONS *
1171              **************************/
1172
1173             /* Compute parameters for interactions between i and j atoms */
1174             qq20             = _mm256_mul_ps(iq2,jq0);
1175
1176             /* COULOMB ELECTROSTATICS */
1177             velec            = _mm256_mul_ps(qq20,rinv20);
1178             felec            = _mm256_mul_ps(velec,rinvsq20);
1179
1180             fscal            = felec;
1181
1182             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1183
1184             /* Calculate temporary vectorial force */
1185             tx               = _mm256_mul_ps(fscal,dx20);
1186             ty               = _mm256_mul_ps(fscal,dy20);
1187             tz               = _mm256_mul_ps(fscal,dz20);
1188
1189             /* Update vectorial force */
1190             fix2             = _mm256_add_ps(fix2,tx);
1191             fiy2             = _mm256_add_ps(fiy2,ty);
1192             fiz2             = _mm256_add_ps(fiz2,tz);
1193
1194             fjx0             = _mm256_add_ps(fjx0,tx);
1195             fjy0             = _mm256_add_ps(fjy0,ty);
1196             fjz0             = _mm256_add_ps(fjz0,tz);
1197
1198             /**************************
1199              * CALCULATE INTERACTIONS *
1200              **************************/
1201
1202             /* Compute parameters for interactions between i and j atoms */
1203             qq30             = _mm256_mul_ps(iq3,jq0);
1204
1205             /* COULOMB ELECTROSTATICS */
1206             velec            = _mm256_mul_ps(qq30,rinv30);
1207             felec            = _mm256_mul_ps(velec,rinvsq30);
1208
1209             fscal            = felec;
1210
1211             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1212
1213             /* Calculate temporary vectorial force */
1214             tx               = _mm256_mul_ps(fscal,dx30);
1215             ty               = _mm256_mul_ps(fscal,dy30);
1216             tz               = _mm256_mul_ps(fscal,dz30);
1217
1218             /* Update vectorial force */
1219             fix3             = _mm256_add_ps(fix3,tx);
1220             fiy3             = _mm256_add_ps(fiy3,ty);
1221             fiz3             = _mm256_add_ps(fiz3,tz);
1222
1223             fjx0             = _mm256_add_ps(fjx0,tx);
1224             fjy0             = _mm256_add_ps(fjy0,ty);
1225             fjz0             = _mm256_add_ps(fjz0,tz);
1226
1227             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1228             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1229             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1230             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1231             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1232             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1233             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1234             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1235
1236             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1237
1238             /* Inner loop uses 108 flops */
1239         }
1240
1241         /* End of innermost loop */
1242
1243         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1244                                                  f+i_coord_offset,fshift+i_shift_offset);
1245
1246         /* Increment number of inner iterations */
1247         inneriter                  += j_index_end - j_index_start;
1248
1249         /* Outer loop uses 24 flops */
1250     }
1251
1252     /* Increment number of outer iterations */
1253     outeriter        += nri;
1254
1255     /* Update outer/inner flops */
1256
1257     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1258 }