src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwLJ_GeomW3P1_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_ElecEw_VdwLJ_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwLJ_GeomW3P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  90     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  91     __m256i          ewitab;
  92     __m128i          ewitab_lo,ewitab_hi;
  93     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  95     real             *ewtab;
  96     __m256           dummy_mask,cutoff_mask;
  97     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  98     __m256           one     = _mm256_set1_ps(1.0);
  99     __m256           two     = _mm256_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm256_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 118     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 119     beta2            = _mm256_mul_ps(beta,beta);
 120     beta3            = _mm256_mul_ps(beta,beta2);
 121
 122     ewtab            = fr->ic->tabq_coul_FDV0;
 123     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 124     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 129     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 130     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 131     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139     j_coord_offsetE = 0;
 140     j_coord_offsetF = 0;
 141     j_coord_offsetG = 0;
 142     j_coord_offsetH = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm256_setzero_ps();
 171         fiy0             = _mm256_setzero_ps();
 172         fiz0             = _mm256_setzero_ps();
 173         fix1             = _mm256_setzero_ps();
 174         fiy1             = _mm256_setzero_ps();
 175         fiz1             = _mm256_setzero_ps();
 176         fix2             = _mm256_setzero_ps();
 177         fiy2             = _mm256_setzero_ps();
 178         fiz2             = _mm256_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm256_setzero_ps();
 182         vvdwsum          = _mm256_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             jnrE             = jjnr[jidx+4];
 194             jnrF             = jjnr[jidx+5];
 195             jnrG             = jjnr[jidx+6];
 196             jnrH             = jjnr[jidx+7];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201             j_coord_offsetE  = DIM*jnrE;
 202             j_coord_offsetF  = DIM*jnrF;
 203             j_coord_offsetG  = DIM*jnrG;
 204             j_coord_offsetH  = DIM*jnrH;
 205
 206             /* load j atom coordinates */
 207             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 209                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 210                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_ps(ix0,jx0);
 215             dy00             = _mm256_sub_ps(iy0,jy0);
 216             dz00             = _mm256_sub_ps(iz0,jz0);
 217             dx10             = _mm256_sub_ps(ix1,jx0);
 218             dy10             = _mm256_sub_ps(iy1,jy0);
 219             dz10             = _mm256_sub_ps(iz1,jz0);
 220             dx20             = _mm256_sub_ps(ix2,jx0);
 221             dy20             = _mm256_sub_ps(iy2,jy0);
 222             dz20             = _mm256_sub_ps(iz2,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 226             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 227             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 228
 229             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 230             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 231             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 232
 233             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 234             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 235             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 239                                                                  charge+jnrC+0,charge+jnrD+0,
 240                                                                  charge+jnrE+0,charge+jnrF+0,
 241                                                                  charge+jnrG+0,charge+jnrH+0);
 242             vdwjidx0A        = 2*vdwtype[jnrA+0];
 243             vdwjidx0B        = 2*vdwtype[jnrB+0];
 244             vdwjidx0C        = 2*vdwtype[jnrC+0];
 245             vdwjidx0D        = 2*vdwtype[jnrD+0];
 246             vdwjidx0E        = 2*vdwtype[jnrE+0];
 247             vdwjidx0F        = 2*vdwtype[jnrF+0];
 248             vdwjidx0G        = 2*vdwtype[jnrG+0];
 249             vdwjidx0H        = 2*vdwtype[jnrH+0];
 250
 251             fjx0             = _mm256_setzero_ps();
 252             fjy0             = _mm256_setzero_ps();
 253             fjz0             = _mm256_setzero_ps();
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             r00              = _mm256_mul_ps(rsq00,rinv00);
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             qq00             = _mm256_mul_ps(iq0,jq0);
 263             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 264                                             vdwioffsetptr0+vdwjidx0B,
 265                                             vdwioffsetptr0+vdwjidx0C,
 266                                             vdwioffsetptr0+vdwjidx0D,
 267                                             vdwioffsetptr0+vdwjidx0E,
 268                                             vdwioffsetptr0+vdwjidx0F,
 269                                             vdwioffsetptr0+vdwjidx0G,
 270                                             vdwioffsetptr0+vdwjidx0H,
 271                                             &c6_00,&c12_00);
 272
 273             /* EWALD ELECTROSTATICS */
 274
 275             /* Analytical PME correction */
 276             zeta2            = _mm256_mul_ps(beta2,rsq00);
 277             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 278             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 279             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 280             felec            = _mm256_mul_ps(qq00,felec);
 281             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 282             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 283             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 284             velec            = _mm256_mul_ps(qq00,velec);
 285
 286             /* LENNARD-JONES DISPERSION/REPULSION */
 287
 288             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 289             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 290             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 291             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 292             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velecsum         = _mm256_add_ps(velecsum,velec);
 296             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 297
 298             fscal            = _mm256_add_ps(felec,fvdw);
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm256_mul_ps(fscal,dx00);
 302             ty               = _mm256_mul_ps(fscal,dy00);
 303             tz               = _mm256_mul_ps(fscal,dz00);
 304
 305             /* Update vectorial force */
 306             fix0             = _mm256_add_ps(fix0,tx);
 307             fiy0             = _mm256_add_ps(fiy0,ty);
 308             fiz0             = _mm256_add_ps(fiz0,tz);
 309
 310             fjx0             = _mm256_add_ps(fjx0,tx);
 311             fjy0             = _mm256_add_ps(fjy0,ty);
 312             fjz0             = _mm256_add_ps(fjz0,tz);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r10              = _mm256_mul_ps(rsq10,rinv10);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm256_mul_ps(iq1,jq0);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Analytical PME correction */
 326             zeta2            = _mm256_mul_ps(beta2,rsq10);
 327             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 328             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 329             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 330             felec            = _mm256_mul_ps(qq10,felec);
 331             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 332             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 333             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 334             velec            = _mm256_mul_ps(qq10,velec);
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velecsum         = _mm256_add_ps(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             /* Calculate temporary vectorial force */
 342             tx               = _mm256_mul_ps(fscal,dx10);
 343             ty               = _mm256_mul_ps(fscal,dy10);
 344             tz               = _mm256_mul_ps(fscal,dz10);
 345
 346             /* Update vectorial force */
 347             fix1             = _mm256_add_ps(fix1,tx);
 348             fiy1             = _mm256_add_ps(fiy1,ty);
 349             fiz1             = _mm256_add_ps(fiz1,tz);
 350
 351             fjx0             = _mm256_add_ps(fjx0,tx);
 352             fjy0             = _mm256_add_ps(fjy0,ty);
 353             fjz0             = _mm256_add_ps(fjz0,tz);
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             r20              = _mm256_mul_ps(rsq20,rinv20);
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq20             = _mm256_mul_ps(iq2,jq0);
 363
 364             /* EWALD ELECTROSTATICS */
 365
 366             /* Analytical PME correction */
 367             zeta2            = _mm256_mul_ps(beta2,rsq20);
 368             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 369             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 370             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 371             felec            = _mm256_mul_ps(qq20,felec);
 372             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 373             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 374             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 375             velec            = _mm256_mul_ps(qq20,velec);
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velecsum         = _mm256_add_ps(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm256_mul_ps(fscal,dx20);
 384             ty               = _mm256_mul_ps(fscal,dy20);
 385             tz               = _mm256_mul_ps(fscal,dz20);
 386
 387             /* Update vectorial force */
 388             fix2             = _mm256_add_ps(fix2,tx);
 389             fiy2             = _mm256_add_ps(fiy2,ty);
 390             fiz2             = _mm256_add_ps(fiz2,tz);
 391
 392             fjx0             = _mm256_add_ps(fjx0,tx);
 393             fjy0             = _mm256_add_ps(fjy0,ty);
 394             fjz0             = _mm256_add_ps(fjz0,tz);
 395
 396             fjptrA             = f+j_coord_offsetA;
 397             fjptrB             = f+j_coord_offsetB;
 398             fjptrC             = f+j_coord_offsetC;
 399             fjptrD             = f+j_coord_offsetD;
 400             fjptrE             = f+j_coord_offsetE;
 401             fjptrF             = f+j_coord_offsetF;
 402             fjptrG             = f+j_coord_offsetG;
 403             fjptrH             = f+j_coord_offsetH;
 404
 405             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 406
 407             /* Inner loop uses 267 flops */
 408         }
 409
 410         if(jidx<j_index_end)
 411         {
 412
 413             /* Get j neighbor index, and coordinate index */
 414             jnrlistA         = jjnr[jidx];
 415             jnrlistB         = jjnr[jidx+1];
 416             jnrlistC         = jjnr[jidx+2];
 417             jnrlistD         = jjnr[jidx+3];
 418             jnrlistE         = jjnr[jidx+4];
 419             jnrlistF         = jjnr[jidx+5];
 420             jnrlistG         = jjnr[jidx+6];
 421             jnrlistH         = jjnr[jidx+7];
 422             /* Sign of each element will be negative for non-real atoms.
 423              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 424              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 425              */
 426             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 427                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 428
 429             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 430             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 431             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 432             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 433             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 434             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 435             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 436             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 437             j_coord_offsetA  = DIM*jnrA;
 438             j_coord_offsetB  = DIM*jnrB;
 439             j_coord_offsetC  = DIM*jnrC;
 440             j_coord_offsetD  = DIM*jnrD;
 441             j_coord_offsetE  = DIM*jnrE;
 442             j_coord_offsetF  = DIM*jnrF;
 443             j_coord_offsetG  = DIM*jnrG;
 444             j_coord_offsetH  = DIM*jnrH;
 445
 446             /* load j atom coordinates */
 447             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 448                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 449                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 450                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 451                                                  &jx0,&jy0,&jz0);
 452
 453             /* Calculate displacement vector */
 454             dx00             = _mm256_sub_ps(ix0,jx0);
 455             dy00             = _mm256_sub_ps(iy0,jy0);
 456             dz00             = _mm256_sub_ps(iz0,jz0);
 457             dx10             = _mm256_sub_ps(ix1,jx0);
 458             dy10             = _mm256_sub_ps(iy1,jy0);
 459             dz10             = _mm256_sub_ps(iz1,jz0);
 460             dx20             = _mm256_sub_ps(ix2,jx0);
 461             dy20             = _mm256_sub_ps(iy2,jy0);
 462             dz20             = _mm256_sub_ps(iz2,jz0);
 463
 464             /* Calculate squared distance and things based on it */
 465             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 466             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 467             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 468
 469             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 470             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 471             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 472
 473             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 474             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 475             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 476
 477             /* Load parameters for j particles */
 478             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 479                                                                  charge+jnrC+0,charge+jnrD+0,
 480                                                                  charge+jnrE+0,charge+jnrF+0,
 481                                                                  charge+jnrG+0,charge+jnrH+0);
 482             vdwjidx0A        = 2*vdwtype[jnrA+0];
 483             vdwjidx0B        = 2*vdwtype[jnrB+0];
 484             vdwjidx0C        = 2*vdwtype[jnrC+0];
 485             vdwjidx0D        = 2*vdwtype[jnrD+0];
 486             vdwjidx0E        = 2*vdwtype[jnrE+0];
 487             vdwjidx0F        = 2*vdwtype[jnrF+0];
 488             vdwjidx0G        = 2*vdwtype[jnrG+0];
 489             vdwjidx0H        = 2*vdwtype[jnrH+0];
 490
 491             fjx0             = _mm256_setzero_ps();
 492             fjy0             = _mm256_setzero_ps();
 493             fjz0             = _mm256_setzero_ps();
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             r00              = _mm256_mul_ps(rsq00,rinv00);
 500             r00              = _mm256_andnot_ps(dummy_mask,r00);
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             qq00             = _mm256_mul_ps(iq0,jq0);
 504             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 505                                             vdwioffsetptr0+vdwjidx0B,
 506                                             vdwioffsetptr0+vdwjidx0C,
 507                                             vdwioffsetptr0+vdwjidx0D,
 508                                             vdwioffsetptr0+vdwjidx0E,
 509                                             vdwioffsetptr0+vdwjidx0F,
 510                                             vdwioffsetptr0+vdwjidx0G,
 511                                             vdwioffsetptr0+vdwjidx0H,
 512                                             &c6_00,&c12_00);
 513
 514             /* EWALD ELECTROSTATICS */
 515
 516             /* Analytical PME correction */
 517             zeta2            = _mm256_mul_ps(beta2,rsq00);
 518             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 519             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 520             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 521             felec            = _mm256_mul_ps(qq00,felec);
 522             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 523             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 524             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 525             velec            = _mm256_mul_ps(qq00,velec);
 526
 527             /* LENNARD-JONES DISPERSION/REPULSION */
 528
 529             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 530             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 531             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 532             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 533             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 534
 535             /* Update potential sum for this i atom from the interaction with this j atom. */
 536             velec            = _mm256_andnot_ps(dummy_mask,velec);
 537             velecsum         = _mm256_add_ps(velecsum,velec);
 538             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 539             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 540
 541             fscal            = _mm256_add_ps(felec,fvdw);
 542
 543             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm256_mul_ps(fscal,dx00);
 547             ty               = _mm256_mul_ps(fscal,dy00);
 548             tz               = _mm256_mul_ps(fscal,dz00);
 549
 550             /* Update vectorial force */
 551             fix0             = _mm256_add_ps(fix0,tx);
 552             fiy0             = _mm256_add_ps(fiy0,ty);
 553             fiz0             = _mm256_add_ps(fiz0,tz);
 554
 555             fjx0             = _mm256_add_ps(fjx0,tx);
 556             fjy0             = _mm256_add_ps(fjy0,ty);
 557             fjz0             = _mm256_add_ps(fjz0,tz);
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             r10              = _mm256_mul_ps(rsq10,rinv10);
 564             r10              = _mm256_andnot_ps(dummy_mask,r10);
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq10             = _mm256_mul_ps(iq1,jq0);
 568
 569             /* EWALD ELECTROSTATICS */
 570
 571             /* Analytical PME correction */
 572             zeta2            = _mm256_mul_ps(beta2,rsq10);
 573             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 574             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 575             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 576             felec            = _mm256_mul_ps(qq10,felec);
 577             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 578             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 579             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 580             velec            = _mm256_mul_ps(qq10,velec);
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm256_andnot_ps(dummy_mask,velec);
 584             velecsum         = _mm256_add_ps(velecsum,velec);
 585
 586             fscal            = felec;
 587
 588             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 589
 590             /* Calculate temporary vectorial force */
 591             tx               = _mm256_mul_ps(fscal,dx10);
 592             ty               = _mm256_mul_ps(fscal,dy10);
 593             tz               = _mm256_mul_ps(fscal,dz10);
 594
 595             /* Update vectorial force */
 596             fix1             = _mm256_add_ps(fix1,tx);
 597             fiy1             = _mm256_add_ps(fiy1,ty);
 598             fiz1             = _mm256_add_ps(fiz1,tz);
 599
 600             fjx0             = _mm256_add_ps(fjx0,tx);
 601             fjy0             = _mm256_add_ps(fjy0,ty);
 602             fjz0             = _mm256_add_ps(fjz0,tz);
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             r20              = _mm256_mul_ps(rsq20,rinv20);
 609             r20              = _mm256_andnot_ps(dummy_mask,r20);
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq20             = _mm256_mul_ps(iq2,jq0);
 613
 614             /* EWALD ELECTROSTATICS */
 615
 616             /* Analytical PME correction */
 617             zeta2            = _mm256_mul_ps(beta2,rsq20);
 618             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 619             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 620             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 621             felec            = _mm256_mul_ps(qq20,felec);
 622             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 623             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 624             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 625             velec            = _mm256_mul_ps(qq20,velec);
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             velec            = _mm256_andnot_ps(dummy_mask,velec);
 629             velecsum         = _mm256_add_ps(velecsum,velec);
 630
 631             fscal            = felec;
 632
 633             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 634
 635             /* Calculate temporary vectorial force */
 636             tx               = _mm256_mul_ps(fscal,dx20);
 637             ty               = _mm256_mul_ps(fscal,dy20);
 638             tz               = _mm256_mul_ps(fscal,dz20);
 639
 640             /* Update vectorial force */
 641             fix2             = _mm256_add_ps(fix2,tx);
 642             fiy2             = _mm256_add_ps(fiy2,ty);
 643             fiz2             = _mm256_add_ps(fiz2,tz);
 644
 645             fjx0             = _mm256_add_ps(fjx0,tx);
 646             fjy0             = _mm256_add_ps(fjy0,ty);
 647             fjz0             = _mm256_add_ps(fjz0,tz);
 648
 649             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 650             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 651             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 652             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 653             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 654             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 655             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 656             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 657
 658             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 659
 660             /* Inner loop uses 270 flops */
 661         }
 662
 663         /* End of innermost loop */
 664
 665         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 666                                                  f+i_coord_offset,fshift+i_shift_offset);
 667
 668         ggid                        = gid[iidx];
 669         /* Update potential energies */
 670         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 671         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 672
 673         /* Increment number of inner iterations */
 674         inneriter                  += j_index_end - j_index_start;
 675
 676         /* Outer loop uses 20 flops */
 677     }
 678
 679     /* Increment number of outer iterations */
 680     outeriter        += nri;
 681
 682     /* Update outer/inner flops */
 683
 684     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*270);
 685 }
 686 /*
 687  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_single
 688  * Electrostatics interaction: Ewald
 689  * VdW interaction:            LennardJones
 690  * Geometry:                   Water3-Particle
 691  * Calculate force/pot:        Force
 692  */
 693 void
 694 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_single
 695                     (t_nblist * gmx_restrict                nlist,
 696                      rvec * gmx_restrict                    xx,
 697                      rvec * gmx_restrict                    ff,
 698                      t_forcerec * gmx_restrict              fr,
 699                      t_mdatoms * gmx_restrict               mdatoms,
 700                      nb_kernel_data_t * gmx_restrict        kernel_data,
 701                      t_nrnb * gmx_restrict                  nrnb)
 702 {
 703     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 704      * just 0 for non-waters.
 705      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 706      * jnr indices corresponding to data put in the four positions in the SIMD register.
 707      */
 708     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 709     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 710     int              jnrA,jnrB,jnrC,jnrD;
 711     int              jnrE,jnrF,jnrG,jnrH;
 712     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 713     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 714     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 715     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 716     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 717     real             rcutoff_scalar;
 718     real             *shiftvec,*fshift,*x,*f;
 719     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 720     real             scratch[4*DIM];
 721     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 722     real *           vdwioffsetptr0;
 723     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 724     real *           vdwioffsetptr1;
 725     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 726     real *           vdwioffsetptr2;
 727     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 728     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 729     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 730     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 731     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 732     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 733     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 734     real             *charge;
 735     int              nvdwtype;
 736     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 737     int              *vdwtype;
 738     real             *vdwparam;
 739     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 740     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 741     __m256i          ewitab;
 742     __m128i          ewitab_lo,ewitab_hi;
 743     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 744     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 745     real             *ewtab;
 746     __m256           dummy_mask,cutoff_mask;
 747     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 748     __m256           one     = _mm256_set1_ps(1.0);
 749     __m256           two     = _mm256_set1_ps(2.0);
 750     x                = xx[0];
 751     f                = ff[0];
 752
 753     nri              = nlist->nri;
 754     iinr             = nlist->iinr;
 755     jindex           = nlist->jindex;
 756     jjnr             = nlist->jjnr;
 757     shiftidx         = nlist->shift;
 758     gid              = nlist->gid;
 759     shiftvec         = fr->shift_vec[0];
 760     fshift           = fr->fshift[0];
 761     facel            = _mm256_set1_ps(fr->epsfac);
 762     charge           = mdatoms->chargeA;
 763     nvdwtype         = fr->ntype;
 764     vdwparam         = fr->nbfp;
 765     vdwtype          = mdatoms->typeA;
 766
 767     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 768     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 769     beta2            = _mm256_mul_ps(beta,beta);
 770     beta3            = _mm256_mul_ps(beta,beta2);
 771
 772     ewtab            = fr->ic->tabq_coul_F;
 773     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 774     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 775
 776     /* Setup water-specific parameters */
 777     inr              = nlist->iinr[0];
 778     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 779     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 780     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 781     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 782
 783     /* Avoid stupid compiler warnings */
 784     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 785     j_coord_offsetA = 0;
 786     j_coord_offsetB = 0;
 787     j_coord_offsetC = 0;
 788     j_coord_offsetD = 0;
 789     j_coord_offsetE = 0;
 790     j_coord_offsetF = 0;
 791     j_coord_offsetG = 0;
 792     j_coord_offsetH = 0;
 793
 794     outeriter        = 0;
 795     inneriter        = 0;
 796
 797     for(iidx=0;iidx<4*DIM;iidx++)
 798     {
 799         scratch[iidx] = 0.0;
 800     }
 801
 802     /* Start outer loop over neighborlists */
 803     for(iidx=0; iidx<nri; iidx++)
 804     {
 805         /* Load shift vector for this list */
 806         i_shift_offset   = DIM*shiftidx[iidx];
 807
 808         /* Load limits for loop over neighbors */
 809         j_index_start    = jindex[iidx];
 810         j_index_end      = jindex[iidx+1];
 811
 812         /* Get outer coordinate index */
 813         inr              = iinr[iidx];
 814         i_coord_offset   = DIM*inr;
 815
 816         /* Load i particle coords and add shift vector */
 817         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 818                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 819
 820         fix0             = _mm256_setzero_ps();
 821         fiy0             = _mm256_setzero_ps();
 822         fiz0             = _mm256_setzero_ps();
 823         fix1             = _mm256_setzero_ps();
 824         fiy1             = _mm256_setzero_ps();
 825         fiz1             = _mm256_setzero_ps();
 826         fix2             = _mm256_setzero_ps();
 827         fiy2             = _mm256_setzero_ps();
 828         fiz2             = _mm256_setzero_ps();
 829
 830         /* Start inner kernel loop */
 831         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 832         {
 833
 834             /* Get j neighbor index, and coordinate index */
 835             jnrA             = jjnr[jidx];
 836             jnrB             = jjnr[jidx+1];
 837             jnrC             = jjnr[jidx+2];
 838             jnrD             = jjnr[jidx+3];
 839             jnrE             = jjnr[jidx+4];
 840             jnrF             = jjnr[jidx+5];
 841             jnrG             = jjnr[jidx+6];
 842             jnrH             = jjnr[jidx+7];
 843             j_coord_offsetA  = DIM*jnrA;
 844             j_coord_offsetB  = DIM*jnrB;
 845             j_coord_offsetC  = DIM*jnrC;
 846             j_coord_offsetD  = DIM*jnrD;
 847             j_coord_offsetE  = DIM*jnrE;
 848             j_coord_offsetF  = DIM*jnrF;
 849             j_coord_offsetG  = DIM*jnrG;
 850             j_coord_offsetH  = DIM*jnrH;
 851
 852             /* load j atom coordinates */
 853             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 854                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 855                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 856                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 857                                                  &jx0,&jy0,&jz0);
 858
 859             /* Calculate displacement vector */
 860             dx00             = _mm256_sub_ps(ix0,jx0);
 861             dy00             = _mm256_sub_ps(iy0,jy0);
 862             dz00             = _mm256_sub_ps(iz0,jz0);
 863             dx10             = _mm256_sub_ps(ix1,jx0);
 864             dy10             = _mm256_sub_ps(iy1,jy0);
 865             dz10             = _mm256_sub_ps(iz1,jz0);
 866             dx20             = _mm256_sub_ps(ix2,jx0);
 867             dy20             = _mm256_sub_ps(iy2,jy0);
 868             dz20             = _mm256_sub_ps(iz2,jz0);
 869
 870             /* Calculate squared distance and things based on it */
 871             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 872             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 873             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 874
 875             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 876             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 877             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 878
 879             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 880             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 881             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 882
 883             /* Load parameters for j particles */
 884             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 885                                                                  charge+jnrC+0,charge+jnrD+0,
 886                                                                  charge+jnrE+0,charge+jnrF+0,
 887                                                                  charge+jnrG+0,charge+jnrH+0);
 888             vdwjidx0A        = 2*vdwtype[jnrA+0];
 889             vdwjidx0B        = 2*vdwtype[jnrB+0];
 890             vdwjidx0C        = 2*vdwtype[jnrC+0];
 891             vdwjidx0D        = 2*vdwtype[jnrD+0];
 892             vdwjidx0E        = 2*vdwtype[jnrE+0];
 893             vdwjidx0F        = 2*vdwtype[jnrF+0];
 894             vdwjidx0G        = 2*vdwtype[jnrG+0];
 895             vdwjidx0H        = 2*vdwtype[jnrH+0];
 896
 897             fjx0             = _mm256_setzero_ps();
 898             fjy0             = _mm256_setzero_ps();
 899             fjz0             = _mm256_setzero_ps();
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             r00              = _mm256_mul_ps(rsq00,rinv00);
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq00             = _mm256_mul_ps(iq0,jq0);
 909             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 910                                             vdwioffsetptr0+vdwjidx0B,
 911                                             vdwioffsetptr0+vdwjidx0C,
 912                                             vdwioffsetptr0+vdwjidx0D,
 913                                             vdwioffsetptr0+vdwjidx0E,
 914                                             vdwioffsetptr0+vdwjidx0F,
 915                                             vdwioffsetptr0+vdwjidx0G,
 916                                             vdwioffsetptr0+vdwjidx0H,
 917                                             &c6_00,&c12_00);
 918
 919             /* EWALD ELECTROSTATICS */
 920
 921             /* Analytical PME correction */
 922             zeta2            = _mm256_mul_ps(beta2,rsq00);
 923             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 924             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 925             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 926             felec            = _mm256_mul_ps(qq00,felec);
 927
 928             /* LENNARD-JONES DISPERSION/REPULSION */
 929
 930             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 931             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 932
 933             fscal            = _mm256_add_ps(felec,fvdw);
 934
 935             /* Calculate temporary vectorial force */
 936             tx               = _mm256_mul_ps(fscal,dx00);
 937             ty               = _mm256_mul_ps(fscal,dy00);
 938             tz               = _mm256_mul_ps(fscal,dz00);
 939
 940             /* Update vectorial force */
 941             fix0             = _mm256_add_ps(fix0,tx);
 942             fiy0             = _mm256_add_ps(fiy0,ty);
 943             fiz0             = _mm256_add_ps(fiz0,tz);
 944
 945             fjx0             = _mm256_add_ps(fjx0,tx);
 946             fjy0             = _mm256_add_ps(fjy0,ty);
 947             fjz0             = _mm256_add_ps(fjz0,tz);
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             r10              = _mm256_mul_ps(rsq10,rinv10);
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq10             = _mm256_mul_ps(iq1,jq0);
 957
 958             /* EWALD ELECTROSTATICS */
 959
 960             /* Analytical PME correction */
 961             zeta2            = _mm256_mul_ps(beta2,rsq10);
 962             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 963             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 964             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 965             felec            = _mm256_mul_ps(qq10,felec);
 966
 967             fscal            = felec;
 968
 969             /* Calculate temporary vectorial force */
 970             tx               = _mm256_mul_ps(fscal,dx10);
 971             ty               = _mm256_mul_ps(fscal,dy10);
 972             tz               = _mm256_mul_ps(fscal,dz10);
 973
 974             /* Update vectorial force */
 975             fix1             = _mm256_add_ps(fix1,tx);
 976             fiy1             = _mm256_add_ps(fiy1,ty);
 977             fiz1             = _mm256_add_ps(fiz1,tz);
 978
 979             fjx0             = _mm256_add_ps(fjx0,tx);
 980             fjy0             = _mm256_add_ps(fjy0,ty);
 981             fjz0             = _mm256_add_ps(fjz0,tz);
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             r20              = _mm256_mul_ps(rsq20,rinv20);
 988
 989             /* Compute parameters for interactions between i and j atoms */
 990             qq20             = _mm256_mul_ps(iq2,jq0);
 991
 992             /* EWALD ELECTROSTATICS */
 993
 994             /* Analytical PME correction */
 995             zeta2            = _mm256_mul_ps(beta2,rsq20);
 996             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 997             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 998             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 999             felec            = _mm256_mul_ps(qq20,felec);
1000
1001             fscal            = felec;
1002
1003             /* Calculate temporary vectorial force */
1004             tx               = _mm256_mul_ps(fscal,dx20);
1005             ty               = _mm256_mul_ps(fscal,dy20);
1006             tz               = _mm256_mul_ps(fscal,dz20);
1007
1008             /* Update vectorial force */
1009             fix2             = _mm256_add_ps(fix2,tx);
1010             fiy2             = _mm256_add_ps(fiy2,ty);
1011             fiz2             = _mm256_add_ps(fiz2,tz);
1012
1013             fjx0             = _mm256_add_ps(fjx0,tx);
1014             fjy0             = _mm256_add_ps(fjy0,ty);
1015             fjz0             = _mm256_add_ps(fjz0,tz);
1016
1017             fjptrA             = f+j_coord_offsetA;
1018             fjptrB             = f+j_coord_offsetB;
1019             fjptrC             = f+j_coord_offsetC;
1020             fjptrD             = f+j_coord_offsetD;
1021             fjptrE             = f+j_coord_offsetE;
1022             fjptrF             = f+j_coord_offsetF;
1023             fjptrG             = f+j_coord_offsetG;
1024             fjptrH             = f+j_coord_offsetH;
1025
1026             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1027
1028             /* Inner loop uses 178 flops */
1029         }
1030
1031         if(jidx<j_index_end)
1032         {
1033
1034             /* Get j neighbor index, and coordinate index */
1035             jnrlistA         = jjnr[jidx];
1036             jnrlistB         = jjnr[jidx+1];
1037             jnrlistC         = jjnr[jidx+2];
1038             jnrlistD         = jjnr[jidx+3];
1039             jnrlistE         = jjnr[jidx+4];
1040             jnrlistF         = jjnr[jidx+5];
1041             jnrlistG         = jjnr[jidx+6];
1042             jnrlistH         = jjnr[jidx+7];
1043             /* Sign of each element will be negative for non-real atoms.
1044              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1045              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1046              */
1047             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1048                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1049
1050             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1051             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1052             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1053             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1054             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1055             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1056             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1057             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1058             j_coord_offsetA  = DIM*jnrA;
1059             j_coord_offsetB  = DIM*jnrB;
1060             j_coord_offsetC  = DIM*jnrC;
1061             j_coord_offsetD  = DIM*jnrD;
1062             j_coord_offsetE  = DIM*jnrE;
1063             j_coord_offsetF  = DIM*jnrF;
1064             j_coord_offsetG  = DIM*jnrG;
1065             j_coord_offsetH  = DIM*jnrH;
1066
1067             /* load j atom coordinates */
1068             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1069                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1070                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1071                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1072                                                  &jx0,&jy0,&jz0);
1073
1074             /* Calculate displacement vector */
1075             dx00             = _mm256_sub_ps(ix0,jx0);
1076             dy00             = _mm256_sub_ps(iy0,jy0);
1077             dz00             = _mm256_sub_ps(iz0,jz0);
1078             dx10             = _mm256_sub_ps(ix1,jx0);
1079             dy10             = _mm256_sub_ps(iy1,jy0);
1080             dz10             = _mm256_sub_ps(iz1,jz0);
1081             dx20             = _mm256_sub_ps(ix2,jx0);
1082             dy20             = _mm256_sub_ps(iy2,jy0);
1083             dz20             = _mm256_sub_ps(iz2,jz0);
1084
1085             /* Calculate squared distance and things based on it */
1086             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1087             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1088             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1089
1090             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1091             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1092             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1093
1094             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1095             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1096             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1097
1098             /* Load parameters for j particles */
1099             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1100                                                                  charge+jnrC+0,charge+jnrD+0,
1101                                                                  charge+jnrE+0,charge+jnrF+0,
1102                                                                  charge+jnrG+0,charge+jnrH+0);
1103             vdwjidx0A        = 2*vdwtype[jnrA+0];
1104             vdwjidx0B        = 2*vdwtype[jnrB+0];
1105             vdwjidx0C        = 2*vdwtype[jnrC+0];
1106             vdwjidx0D        = 2*vdwtype[jnrD+0];
1107             vdwjidx0E        = 2*vdwtype[jnrE+0];
1108             vdwjidx0F        = 2*vdwtype[jnrF+0];
1109             vdwjidx0G        = 2*vdwtype[jnrG+0];
1110             vdwjidx0H        = 2*vdwtype[jnrH+0];
1111
1112             fjx0             = _mm256_setzero_ps();
1113             fjy0             = _mm256_setzero_ps();
1114             fjz0             = _mm256_setzero_ps();
1115
1116             /**************************
1117              * CALCULATE INTERACTIONS *
1118              **************************/
1119
1120             r00              = _mm256_mul_ps(rsq00,rinv00);
1121             r00              = _mm256_andnot_ps(dummy_mask,r00);
1122
1123             /* Compute parameters for interactions between i and j atoms */
1124             qq00             = _mm256_mul_ps(iq0,jq0);
1125             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1126                                             vdwioffsetptr0+vdwjidx0B,
1127                                             vdwioffsetptr0+vdwjidx0C,
1128                                             vdwioffsetptr0+vdwjidx0D,
1129                                             vdwioffsetptr0+vdwjidx0E,
1130                                             vdwioffsetptr0+vdwjidx0F,
1131                                             vdwioffsetptr0+vdwjidx0G,
1132                                             vdwioffsetptr0+vdwjidx0H,
1133                                             &c6_00,&c12_00);
1134
1135             /* EWALD ELECTROSTATICS */
1136
1137             /* Analytical PME correction */
1138             zeta2            = _mm256_mul_ps(beta2,rsq00);
1139             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1140             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1141             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1142             felec            = _mm256_mul_ps(qq00,felec);
1143
1144             /* LENNARD-JONES DISPERSION/REPULSION */
1145
1146             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1147             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1148
1149             fscal            = _mm256_add_ps(felec,fvdw);
1150
1151             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1152
1153             /* Calculate temporary vectorial force */
1154             tx               = _mm256_mul_ps(fscal,dx00);
1155             ty               = _mm256_mul_ps(fscal,dy00);
1156             tz               = _mm256_mul_ps(fscal,dz00);
1157
1158             /* Update vectorial force */
1159             fix0             = _mm256_add_ps(fix0,tx);
1160             fiy0             = _mm256_add_ps(fiy0,ty);
1161             fiz0             = _mm256_add_ps(fiz0,tz);
1162
1163             fjx0             = _mm256_add_ps(fjx0,tx);
1164             fjy0             = _mm256_add_ps(fjy0,ty);
1165             fjz0             = _mm256_add_ps(fjz0,tz);
1166
1167             /**************************
1168              * CALCULATE INTERACTIONS *
1169              **************************/
1170
1171             r10              = _mm256_mul_ps(rsq10,rinv10);
1172             r10              = _mm256_andnot_ps(dummy_mask,r10);
1173
1174             /* Compute parameters for interactions between i and j atoms */
1175             qq10             = _mm256_mul_ps(iq1,jq0);
1176
1177             /* EWALD ELECTROSTATICS */
1178
1179             /* Analytical PME correction */
1180             zeta2            = _mm256_mul_ps(beta2,rsq10);
1181             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1182             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1183             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1184             felec            = _mm256_mul_ps(qq10,felec);
1185
1186             fscal            = felec;
1187
1188             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1189
1190             /* Calculate temporary vectorial force */
1191             tx               = _mm256_mul_ps(fscal,dx10);
1192             ty               = _mm256_mul_ps(fscal,dy10);
1193             tz               = _mm256_mul_ps(fscal,dz10);
1194
1195             /* Update vectorial force */
1196             fix1             = _mm256_add_ps(fix1,tx);
1197             fiy1             = _mm256_add_ps(fiy1,ty);
1198             fiz1             = _mm256_add_ps(fiz1,tz);
1199
1200             fjx0             = _mm256_add_ps(fjx0,tx);
1201             fjy0             = _mm256_add_ps(fjy0,ty);
1202             fjz0             = _mm256_add_ps(fjz0,tz);
1203
1204             /**************************
1205              * CALCULATE INTERACTIONS *
1206              **************************/
1207
1208             r20              = _mm256_mul_ps(rsq20,rinv20);
1209             r20              = _mm256_andnot_ps(dummy_mask,r20);
1210
1211             /* Compute parameters for interactions between i and j atoms */
1212             qq20             = _mm256_mul_ps(iq2,jq0);
1213
1214             /* EWALD ELECTROSTATICS */
1215
1216             /* Analytical PME correction */
1217             zeta2            = _mm256_mul_ps(beta2,rsq20);
1218             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1219             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1220             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1221             felec            = _mm256_mul_ps(qq20,felec);
1222
1223             fscal            = felec;
1224
1225             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1226
1227             /* Calculate temporary vectorial force */
1228             tx               = _mm256_mul_ps(fscal,dx20);
1229             ty               = _mm256_mul_ps(fscal,dy20);
1230             tz               = _mm256_mul_ps(fscal,dz20);
1231
1232             /* Update vectorial force */
1233             fix2             = _mm256_add_ps(fix2,tx);
1234             fiy2             = _mm256_add_ps(fiy2,ty);
1235             fiz2             = _mm256_add_ps(fiz2,tz);
1236
1237             fjx0             = _mm256_add_ps(fjx0,tx);
1238             fjy0             = _mm256_add_ps(fjy0,ty);
1239             fjz0             = _mm256_add_ps(fjz0,tz);
1240
1241             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1242             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1243             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1244             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1245             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1246             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1247             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1248             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1249
1250             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1251
1252             /* Inner loop uses 181 flops */
1253         }
1254
1255         /* End of innermost loop */
1256
1257         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1258                                                  f+i_coord_offset,fshift+i_shift_offset);
1259
1260         /* Increment number of inner iterations */
1261         inneriter                  += j_index_end - j_index_start;
1262
1263         /* Outer loop uses 18 flops */
1264     }
1265
1266     /* Increment number of outer iterations */
1267     outeriter        += nri;
1268
1269     /* Update outer/inner flops */
1270
1271     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*181);
1272 }