src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwLJ_GeomW3P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double 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_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_256_double
  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_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  79     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  80     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  81     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  82     real             *charge;
  83     int              nvdwtype;
  84     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  85     int              *vdwtype;
  86     real             *vdwparam;
  87     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  88     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  89     __m128i          ewitab;
  90     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  91     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  92     real             *ewtab;
  93     __m256d          dummy_mask,cutoff_mask;
  94     __m128           tmpmask0,tmpmask1;
  95     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  96     __m256d          one     = _mm256_set1_pd(1.0);
  97     __m256d          two     = _mm256_set1_pd(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_pd(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 116     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 117     beta2            = _mm256_mul_pd(beta,beta);
 118     beta3            = _mm256_mul_pd(beta,beta2);
 119
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 127     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 128     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 129     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 163
 164         fix0             = _mm256_setzero_pd();
 165         fiy0             = _mm256_setzero_pd();
 166         fiz0             = _mm256_setzero_pd();
 167         fix1             = _mm256_setzero_pd();
 168         fiy1             = _mm256_setzero_pd();
 169         fiz1             = _mm256_setzero_pd();
 170         fix2             = _mm256_setzero_pd();
 171         fiy2             = _mm256_setzero_pd();
 172         fiz2             = _mm256_setzero_pd();
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm256_setzero_pd();
 176         vvdwsum          = _mm256_setzero_pd();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 195                                                  &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm256_sub_pd(ix0,jx0);
 199             dy00             = _mm256_sub_pd(iy0,jy0);
 200             dz00             = _mm256_sub_pd(iz0,jz0);
 201             dx10             = _mm256_sub_pd(ix1,jx0);
 202             dy10             = _mm256_sub_pd(iy1,jy0);
 203             dz10             = _mm256_sub_pd(iz1,jz0);
 204             dx20             = _mm256_sub_pd(ix2,jx0);
 205             dy20             = _mm256_sub_pd(iy2,jy0);
 206             dz20             = _mm256_sub_pd(iz2,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 210             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 211             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 212
 213             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 214             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 215             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 216
 217             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 218             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 219             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 223                                                                  charge+jnrC+0,charge+jnrD+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228
 229             fjx0             = _mm256_setzero_pd();
 230             fjy0             = _mm256_setzero_pd();
 231             fjz0             = _mm256_setzero_pd();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             r00              = _mm256_mul_pd(rsq00,rinv00);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _mm256_mul_pd(iq0,jq0);
 241             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 242                                             vdwioffsetptr0+vdwjidx0B,
 243                                             vdwioffsetptr0+vdwjidx0C,
 244                                             vdwioffsetptr0+vdwjidx0D,
 245                                             &c6_00,&c12_00);
 246
 247             /* EWALD ELECTROSTATICS */
 248
 249             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 250             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 251             ewitab           = _mm256_cvttpd_epi32(ewrt);
 252             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 253             ewitab           = _mm_slli_epi32(ewitab,2);
 254             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 255             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 256             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 257             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 258             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 259             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 260             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 261             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 262             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 263
 264             /* LENNARD-JONES DISPERSION/REPULSION */
 265
 266             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 267             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 268             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 269             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 270             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velecsum         = _mm256_add_pd(velecsum,velec);
 274             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 275
 276             fscal            = _mm256_add_pd(felec,fvdw);
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm256_mul_pd(fscal,dx00);
 280             ty               = _mm256_mul_pd(fscal,dy00);
 281             tz               = _mm256_mul_pd(fscal,dz00);
 282
 283             /* Update vectorial force */
 284             fix0             = _mm256_add_pd(fix0,tx);
 285             fiy0             = _mm256_add_pd(fiy0,ty);
 286             fiz0             = _mm256_add_pd(fiz0,tz);
 287
 288             fjx0             = _mm256_add_pd(fjx0,tx);
 289             fjy0             = _mm256_add_pd(fjy0,ty);
 290             fjz0             = _mm256_add_pd(fjz0,tz);
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             r10              = _mm256_mul_pd(rsq10,rinv10);
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq10             = _mm256_mul_pd(iq1,jq0);
 300
 301             /* EWALD ELECTROSTATICS */
 302
 303             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 304             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 305             ewitab           = _mm256_cvttpd_epi32(ewrt);
 306             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 307             ewitab           = _mm_slli_epi32(ewitab,2);
 308             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 309             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 310             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 311             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 312             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 313             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 314             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 315             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 316             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velecsum         = _mm256_add_pd(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm256_mul_pd(fscal,dx10);
 325             ty               = _mm256_mul_pd(fscal,dy10);
 326             tz               = _mm256_mul_pd(fscal,dz10);
 327
 328             /* Update vectorial force */
 329             fix1             = _mm256_add_pd(fix1,tx);
 330             fiy1             = _mm256_add_pd(fiy1,ty);
 331             fiz1             = _mm256_add_pd(fiz1,tz);
 332
 333             fjx0             = _mm256_add_pd(fjx0,tx);
 334             fjy0             = _mm256_add_pd(fjy0,ty);
 335             fjz0             = _mm256_add_pd(fjz0,tz);
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             r20              = _mm256_mul_pd(rsq20,rinv20);
 342
 343             /* Compute parameters for interactions between i and j atoms */
 344             qq20             = _mm256_mul_pd(iq2,jq0);
 345
 346             /* EWALD ELECTROSTATICS */
 347
 348             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 349             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 350             ewitab           = _mm256_cvttpd_epi32(ewrt);
 351             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 352             ewitab           = _mm_slli_epi32(ewitab,2);
 353             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 354             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 355             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 356             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 357             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 358             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 359             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 360             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 361             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velecsum         = _mm256_add_pd(velecsum,velec);
 365
 366             fscal            = felec;
 367
 368             /* Calculate temporary vectorial force */
 369             tx               = _mm256_mul_pd(fscal,dx20);
 370             ty               = _mm256_mul_pd(fscal,dy20);
 371             tz               = _mm256_mul_pd(fscal,dz20);
 372
 373             /* Update vectorial force */
 374             fix2             = _mm256_add_pd(fix2,tx);
 375             fiy2             = _mm256_add_pd(fiy2,ty);
 376             fiz2             = _mm256_add_pd(fiz2,tz);
 377
 378             fjx0             = _mm256_add_pd(fjx0,tx);
 379             fjy0             = _mm256_add_pd(fjy0,ty);
 380             fjz0             = _mm256_add_pd(fjz0,tz);
 381
 382             fjptrA             = f+j_coord_offsetA;
 383             fjptrB             = f+j_coord_offsetB;
 384             fjptrC             = f+j_coord_offsetC;
 385             fjptrD             = f+j_coord_offsetD;
 386
 387             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 388
 389             /* Inner loop uses 138 flops */
 390         }
 391
 392         if(jidx<j_index_end)
 393         {
 394
 395             /* Get j neighbor index, and coordinate index */
 396             jnrlistA         = jjnr[jidx];
 397             jnrlistB         = jjnr[jidx+1];
 398             jnrlistC         = jjnr[jidx+2];
 399             jnrlistD         = jjnr[jidx+3];
 400             /* Sign of each element will be negative for non-real atoms.
 401              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 402              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 403              */
 404             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 405
 406             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 407             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 408             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 409
 410             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 411             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 412             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 413             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 414             j_coord_offsetA  = DIM*jnrA;
 415             j_coord_offsetB  = DIM*jnrB;
 416             j_coord_offsetC  = DIM*jnrC;
 417             j_coord_offsetD  = DIM*jnrD;
 418
 419             /* load j atom coordinates */
 420             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 421                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 422                                                  &jx0,&jy0,&jz0);
 423
 424             /* Calculate displacement vector */
 425             dx00             = _mm256_sub_pd(ix0,jx0);
 426             dy00             = _mm256_sub_pd(iy0,jy0);
 427             dz00             = _mm256_sub_pd(iz0,jz0);
 428             dx10             = _mm256_sub_pd(ix1,jx0);
 429             dy10             = _mm256_sub_pd(iy1,jy0);
 430             dz10             = _mm256_sub_pd(iz1,jz0);
 431             dx20             = _mm256_sub_pd(ix2,jx0);
 432             dy20             = _mm256_sub_pd(iy2,jy0);
 433             dz20             = _mm256_sub_pd(iz2,jz0);
 434
 435             /* Calculate squared distance and things based on it */
 436             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 437             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 438             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 439
 440             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 441             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 442             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 443
 444             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 445             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 446             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 447
 448             /* Load parameters for j particles */
 449             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 450                                                                  charge+jnrC+0,charge+jnrD+0);
 451             vdwjidx0A        = 2*vdwtype[jnrA+0];
 452             vdwjidx0B        = 2*vdwtype[jnrB+0];
 453             vdwjidx0C        = 2*vdwtype[jnrC+0];
 454             vdwjidx0D        = 2*vdwtype[jnrD+0];
 455
 456             fjx0             = _mm256_setzero_pd();
 457             fjy0             = _mm256_setzero_pd();
 458             fjz0             = _mm256_setzero_pd();
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             r00              = _mm256_mul_pd(rsq00,rinv00);
 465             r00              = _mm256_andnot_pd(dummy_mask,r00);
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             qq00             = _mm256_mul_pd(iq0,jq0);
 469             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 470                                             vdwioffsetptr0+vdwjidx0B,
 471                                             vdwioffsetptr0+vdwjidx0C,
 472                                             vdwioffsetptr0+vdwjidx0D,
 473                                             &c6_00,&c12_00);
 474
 475             /* EWALD ELECTROSTATICS */
 476
 477             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 478             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 479             ewitab           = _mm256_cvttpd_epi32(ewrt);
 480             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 481             ewitab           = _mm_slli_epi32(ewitab,2);
 482             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 483             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 484             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 485             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 486             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 487             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 488             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 489             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 490             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 491
 492             /* LENNARD-JONES DISPERSION/REPULSION */
 493
 494             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 495             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 496             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 497             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 498             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 499
 500             /* Update potential sum for this i atom from the interaction with this j atom. */
 501             velec            = _mm256_andnot_pd(dummy_mask,velec);
 502             velecsum         = _mm256_add_pd(velecsum,velec);
 503             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 504             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 505
 506             fscal            = _mm256_add_pd(felec,fvdw);
 507
 508             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 509
 510             /* Calculate temporary vectorial force */
 511             tx               = _mm256_mul_pd(fscal,dx00);
 512             ty               = _mm256_mul_pd(fscal,dy00);
 513             tz               = _mm256_mul_pd(fscal,dz00);
 514
 515             /* Update vectorial force */
 516             fix0             = _mm256_add_pd(fix0,tx);
 517             fiy0             = _mm256_add_pd(fiy0,ty);
 518             fiz0             = _mm256_add_pd(fiz0,tz);
 519
 520             fjx0             = _mm256_add_pd(fjx0,tx);
 521             fjy0             = _mm256_add_pd(fjy0,ty);
 522             fjz0             = _mm256_add_pd(fjz0,tz);
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r10              = _mm256_mul_pd(rsq10,rinv10);
 529             r10              = _mm256_andnot_pd(dummy_mask,r10);
 530
 531             /* Compute parameters for interactions between i and j atoms */
 532             qq10             = _mm256_mul_pd(iq1,jq0);
 533
 534             /* EWALD ELECTROSTATICS */
 535
 536             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 537             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 538             ewitab           = _mm256_cvttpd_epi32(ewrt);
 539             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 540             ewitab           = _mm_slli_epi32(ewitab,2);
 541             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 542             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 543             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 544             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 545             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 546             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 547             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 548             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 549             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm256_andnot_pd(dummy_mask,velec);
 553             velecsum         = _mm256_add_pd(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm256_mul_pd(fscal,dx10);
 561             ty               = _mm256_mul_pd(fscal,dy10);
 562             tz               = _mm256_mul_pd(fscal,dz10);
 563
 564             /* Update vectorial force */
 565             fix1             = _mm256_add_pd(fix1,tx);
 566             fiy1             = _mm256_add_pd(fiy1,ty);
 567             fiz1             = _mm256_add_pd(fiz1,tz);
 568
 569             fjx0             = _mm256_add_pd(fjx0,tx);
 570             fjy0             = _mm256_add_pd(fjy0,ty);
 571             fjz0             = _mm256_add_pd(fjz0,tz);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             r20              = _mm256_mul_pd(rsq20,rinv20);
 578             r20              = _mm256_andnot_pd(dummy_mask,r20);
 579
 580             /* Compute parameters for interactions between i and j atoms */
 581             qq20             = _mm256_mul_pd(iq2,jq0);
 582
 583             /* EWALD ELECTROSTATICS */
 584
 585             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 586             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 587             ewitab           = _mm256_cvttpd_epi32(ewrt);
 588             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 589             ewitab           = _mm_slli_epi32(ewitab,2);
 590             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 591             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 592             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 593             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 594             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 595             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 596             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 597             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 598             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 599
 600             /* Update potential sum for this i atom from the interaction with this j atom. */
 601             velec            = _mm256_andnot_pd(dummy_mask,velec);
 602             velecsum         = _mm256_add_pd(velecsum,velec);
 603
 604             fscal            = felec;
 605
 606             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 607
 608             /* Calculate temporary vectorial force */
 609             tx               = _mm256_mul_pd(fscal,dx20);
 610             ty               = _mm256_mul_pd(fscal,dy20);
 611             tz               = _mm256_mul_pd(fscal,dz20);
 612
 613             /* Update vectorial force */
 614             fix2             = _mm256_add_pd(fix2,tx);
 615             fiy2             = _mm256_add_pd(fiy2,ty);
 616             fiz2             = _mm256_add_pd(fiz2,tz);
 617
 618             fjx0             = _mm256_add_pd(fjx0,tx);
 619             fjy0             = _mm256_add_pd(fjy0,ty);
 620             fjz0             = _mm256_add_pd(fjz0,tz);
 621
 622             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 623             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 624             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 625             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 626
 627             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 628
 629             /* Inner loop uses 141 flops */
 630         }
 631
 632         /* End of innermost loop */
 633
 634         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 635                                                  f+i_coord_offset,fshift+i_shift_offset);
 636
 637         ggid                        = gid[iidx];
 638         /* Update potential energies */
 639         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 640         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 641
 642         /* Increment number of inner iterations */
 643         inneriter                  += j_index_end - j_index_start;
 644
 645         /* Outer loop uses 20 flops */
 646     }
 647
 648     /* Increment number of outer iterations */
 649     outeriter        += nri;
 650
 651     /* Update outer/inner flops */
 652
 653     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*141);
 654 }
 655 /*
 656  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_double
 657  * Electrostatics interaction: Ewald
 658  * VdW interaction:            LennardJones
 659  * Geometry:                   Water3-Particle
 660  * Calculate force/pot:        Force
 661  */
 662 void
 663 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_double
 664                     (t_nblist * gmx_restrict                nlist,
 665                      rvec * gmx_restrict                    xx,
 666                      rvec * gmx_restrict                    ff,
 667                      t_forcerec * gmx_restrict              fr,
 668                      t_mdatoms * gmx_restrict               mdatoms,
 669                      nb_kernel_data_t * gmx_restrict        kernel_data,
 670                      t_nrnb * gmx_restrict                  nrnb)
 671 {
 672     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 673      * just 0 for non-waters.
 674      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 675      * jnr indices corresponding to data put in the four positions in the SIMD register.
 676      */
 677     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 678     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 679     int              jnrA,jnrB,jnrC,jnrD;
 680     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 681     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 682     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 683     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 684     real             rcutoff_scalar;
 685     real             *shiftvec,*fshift,*x,*f;
 686     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 687     real             scratch[4*DIM];
 688     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 689     real *           vdwioffsetptr0;
 690     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 691     real *           vdwioffsetptr1;
 692     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 693     real *           vdwioffsetptr2;
 694     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 695     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 696     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 697     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 698     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 699     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 700     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 701     real             *charge;
 702     int              nvdwtype;
 703     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 704     int              *vdwtype;
 705     real             *vdwparam;
 706     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 707     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 708     __m128i          ewitab;
 709     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 710     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 711     real             *ewtab;
 712     __m256d          dummy_mask,cutoff_mask;
 713     __m128           tmpmask0,tmpmask1;
 714     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 715     __m256d          one     = _mm256_set1_pd(1.0);
 716     __m256d          two     = _mm256_set1_pd(2.0);
 717     x                = xx[0];
 718     f                = ff[0];
 719
 720     nri              = nlist->nri;
 721     iinr             = nlist->iinr;
 722     jindex           = nlist->jindex;
 723     jjnr             = nlist->jjnr;
 724     shiftidx         = nlist->shift;
 725     gid              = nlist->gid;
 726     shiftvec         = fr->shift_vec[0];
 727     fshift           = fr->fshift[0];
 728     facel            = _mm256_set1_pd(fr->epsfac);
 729     charge           = mdatoms->chargeA;
 730     nvdwtype         = fr->ntype;
 731     vdwparam         = fr->nbfp;
 732     vdwtype          = mdatoms->typeA;
 733
 734     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 735     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 736     beta2            = _mm256_mul_pd(beta,beta);
 737     beta3            = _mm256_mul_pd(beta,beta2);
 738
 739     ewtab            = fr->ic->tabq_coul_F;
 740     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 741     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 742
 743     /* Setup water-specific parameters */
 744     inr              = nlist->iinr[0];
 745     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 746     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 747     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 748     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 749
 750     /* Avoid stupid compiler warnings */
 751     jnrA = jnrB = jnrC = jnrD = 0;
 752     j_coord_offsetA = 0;
 753     j_coord_offsetB = 0;
 754     j_coord_offsetC = 0;
 755     j_coord_offsetD = 0;
 756
 757     outeriter        = 0;
 758     inneriter        = 0;
 759
 760     for(iidx=0;iidx<4*DIM;iidx++)
 761     {
 762         scratch[iidx] = 0.0;
 763     }
 764
 765     /* Start outer loop over neighborlists */
 766     for(iidx=0; iidx<nri; iidx++)
 767     {
 768         /* Load shift vector for this list */
 769         i_shift_offset   = DIM*shiftidx[iidx];
 770
 771         /* Load limits for loop over neighbors */
 772         j_index_start    = jindex[iidx];
 773         j_index_end      = jindex[iidx+1];
 774
 775         /* Get outer coordinate index */
 776         inr              = iinr[iidx];
 777         i_coord_offset   = DIM*inr;
 778
 779         /* Load i particle coords and add shift vector */
 780         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 781                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 782
 783         fix0             = _mm256_setzero_pd();
 784         fiy0             = _mm256_setzero_pd();
 785         fiz0             = _mm256_setzero_pd();
 786         fix1             = _mm256_setzero_pd();
 787         fiy1             = _mm256_setzero_pd();
 788         fiz1             = _mm256_setzero_pd();
 789         fix2             = _mm256_setzero_pd();
 790         fiy2             = _mm256_setzero_pd();
 791         fiz2             = _mm256_setzero_pd();
 792
 793         /* Start inner kernel loop */
 794         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 795         {
 796
 797             /* Get j neighbor index, and coordinate index */
 798             jnrA             = jjnr[jidx];
 799             jnrB             = jjnr[jidx+1];
 800             jnrC             = jjnr[jidx+2];
 801             jnrD             = jjnr[jidx+3];
 802             j_coord_offsetA  = DIM*jnrA;
 803             j_coord_offsetB  = DIM*jnrB;
 804             j_coord_offsetC  = DIM*jnrC;
 805             j_coord_offsetD  = DIM*jnrD;
 806
 807             /* load j atom coordinates */
 808             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 809                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 810                                                  &jx0,&jy0,&jz0);
 811
 812             /* Calculate displacement vector */
 813             dx00             = _mm256_sub_pd(ix0,jx0);
 814             dy00             = _mm256_sub_pd(iy0,jy0);
 815             dz00             = _mm256_sub_pd(iz0,jz0);
 816             dx10             = _mm256_sub_pd(ix1,jx0);
 817             dy10             = _mm256_sub_pd(iy1,jy0);
 818             dz10             = _mm256_sub_pd(iz1,jz0);
 819             dx20             = _mm256_sub_pd(ix2,jx0);
 820             dy20             = _mm256_sub_pd(iy2,jy0);
 821             dz20             = _mm256_sub_pd(iz2,jz0);
 822
 823             /* Calculate squared distance and things based on it */
 824             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 825             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 826             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 827
 828             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 829             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 830             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 831
 832             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 833             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 834             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 835
 836             /* Load parameters for j particles */
 837             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 838                                                                  charge+jnrC+0,charge+jnrD+0);
 839             vdwjidx0A        = 2*vdwtype[jnrA+0];
 840             vdwjidx0B        = 2*vdwtype[jnrB+0];
 841             vdwjidx0C        = 2*vdwtype[jnrC+0];
 842             vdwjidx0D        = 2*vdwtype[jnrD+0];
 843
 844             fjx0             = _mm256_setzero_pd();
 845             fjy0             = _mm256_setzero_pd();
 846             fjz0             = _mm256_setzero_pd();
 847
 848             /**************************
 849              * CALCULATE INTERACTIONS *
 850              **************************/
 851
 852             r00              = _mm256_mul_pd(rsq00,rinv00);
 853
 854             /* Compute parameters for interactions between i and j atoms */
 855             qq00             = _mm256_mul_pd(iq0,jq0);
 856             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 857                                             vdwioffsetptr0+vdwjidx0B,
 858                                             vdwioffsetptr0+vdwjidx0C,
 859                                             vdwioffsetptr0+vdwjidx0D,
 860                                             &c6_00,&c12_00);
 861
 862             /* EWALD ELECTROSTATICS */
 863
 864             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 865             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 866             ewitab           = _mm256_cvttpd_epi32(ewrt);
 867             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 868             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 869                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 870                                             &ewtabF,&ewtabFn);
 871             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 872             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 873
 874             /* LENNARD-JONES DISPERSION/REPULSION */
 875
 876             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 877             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 878
 879             fscal            = _mm256_add_pd(felec,fvdw);
 880
 881             /* Calculate temporary vectorial force */
 882             tx               = _mm256_mul_pd(fscal,dx00);
 883             ty               = _mm256_mul_pd(fscal,dy00);
 884             tz               = _mm256_mul_pd(fscal,dz00);
 885
 886             /* Update vectorial force */
 887             fix0             = _mm256_add_pd(fix0,tx);
 888             fiy0             = _mm256_add_pd(fiy0,ty);
 889             fiz0             = _mm256_add_pd(fiz0,tz);
 890
 891             fjx0             = _mm256_add_pd(fjx0,tx);
 892             fjy0             = _mm256_add_pd(fjy0,ty);
 893             fjz0             = _mm256_add_pd(fjz0,tz);
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             r10              = _mm256_mul_pd(rsq10,rinv10);
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq10             = _mm256_mul_pd(iq1,jq0);
 903
 904             /* EWALD ELECTROSTATICS */
 905
 906             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 907             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 908             ewitab           = _mm256_cvttpd_epi32(ewrt);
 909             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 910             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 911                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 912                                             &ewtabF,&ewtabFn);
 913             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 914             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 915
 916             fscal            = felec;
 917
 918             /* Calculate temporary vectorial force */
 919             tx               = _mm256_mul_pd(fscal,dx10);
 920             ty               = _mm256_mul_pd(fscal,dy10);
 921             tz               = _mm256_mul_pd(fscal,dz10);
 922
 923             /* Update vectorial force */
 924             fix1             = _mm256_add_pd(fix1,tx);
 925             fiy1             = _mm256_add_pd(fiy1,ty);
 926             fiz1             = _mm256_add_pd(fiz1,tz);
 927
 928             fjx0             = _mm256_add_pd(fjx0,tx);
 929             fjy0             = _mm256_add_pd(fjy0,ty);
 930             fjz0             = _mm256_add_pd(fjz0,tz);
 931
 932             /**************************
 933              * CALCULATE INTERACTIONS *
 934              **************************/
 935
 936             r20              = _mm256_mul_pd(rsq20,rinv20);
 937
 938             /* Compute parameters for interactions between i and j atoms */
 939             qq20             = _mm256_mul_pd(iq2,jq0);
 940
 941             /* EWALD ELECTROSTATICS */
 942
 943             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 944             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 945             ewitab           = _mm256_cvttpd_epi32(ewrt);
 946             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 947             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 948                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 949                                             &ewtabF,&ewtabFn);
 950             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 951             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 952
 953             fscal            = felec;
 954
 955             /* Calculate temporary vectorial force */
 956             tx               = _mm256_mul_pd(fscal,dx20);
 957             ty               = _mm256_mul_pd(fscal,dy20);
 958             tz               = _mm256_mul_pd(fscal,dz20);
 959
 960             /* Update vectorial force */
 961             fix2             = _mm256_add_pd(fix2,tx);
 962             fiy2             = _mm256_add_pd(fiy2,ty);
 963             fiz2             = _mm256_add_pd(fiz2,tz);
 964
 965             fjx0             = _mm256_add_pd(fjx0,tx);
 966             fjy0             = _mm256_add_pd(fjy0,ty);
 967             fjz0             = _mm256_add_pd(fjz0,tz);
 968
 969             fjptrA             = f+j_coord_offsetA;
 970             fjptrB             = f+j_coord_offsetB;
 971             fjptrC             = f+j_coord_offsetC;
 972             fjptrD             = f+j_coord_offsetD;
 973
 974             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 975
 976             /* Inner loop uses 118 flops */
 977         }
 978
 979         if(jidx<j_index_end)
 980         {
 981
 982             /* Get j neighbor index, and coordinate index */
 983             jnrlistA         = jjnr[jidx];
 984             jnrlistB         = jjnr[jidx+1];
 985             jnrlistC         = jjnr[jidx+2];
 986             jnrlistD         = jjnr[jidx+3];
 987             /* Sign of each element will be negative for non-real atoms.
 988              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 989              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 990              */
 991             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 992
 993             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 994             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 995             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 996
 997             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 998             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 999             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1000             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1001             j_coord_offsetA  = DIM*jnrA;
1002             j_coord_offsetB  = DIM*jnrB;
1003             j_coord_offsetC  = DIM*jnrC;
1004             j_coord_offsetD  = DIM*jnrD;
1005
1006             /* load j atom coordinates */
1007             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1008                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1009                                                  &jx0,&jy0,&jz0);
1010
1011             /* Calculate displacement vector */
1012             dx00             = _mm256_sub_pd(ix0,jx0);
1013             dy00             = _mm256_sub_pd(iy0,jy0);
1014             dz00             = _mm256_sub_pd(iz0,jz0);
1015             dx10             = _mm256_sub_pd(ix1,jx0);
1016             dy10             = _mm256_sub_pd(iy1,jy0);
1017             dz10             = _mm256_sub_pd(iz1,jz0);
1018             dx20             = _mm256_sub_pd(ix2,jx0);
1019             dy20             = _mm256_sub_pd(iy2,jy0);
1020             dz20             = _mm256_sub_pd(iz2,jz0);
1021
1022             /* Calculate squared distance and things based on it */
1023             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1024             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1025             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1026
1027             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1028             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1029             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1030
1031             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1032             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1033             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1034
1035             /* Load parameters for j particles */
1036             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1037                                                                  charge+jnrC+0,charge+jnrD+0);
1038             vdwjidx0A        = 2*vdwtype[jnrA+0];
1039             vdwjidx0B        = 2*vdwtype[jnrB+0];
1040             vdwjidx0C        = 2*vdwtype[jnrC+0];
1041             vdwjidx0D        = 2*vdwtype[jnrD+0];
1042
1043             fjx0             = _mm256_setzero_pd();
1044             fjy0             = _mm256_setzero_pd();
1045             fjz0             = _mm256_setzero_pd();
1046
1047             /**************************
1048              * CALCULATE INTERACTIONS *
1049              **************************/
1050
1051             r00              = _mm256_mul_pd(rsq00,rinv00);
1052             r00              = _mm256_andnot_pd(dummy_mask,r00);
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             qq00             = _mm256_mul_pd(iq0,jq0);
1056             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1057                                             vdwioffsetptr0+vdwjidx0B,
1058                                             vdwioffsetptr0+vdwjidx0C,
1059                                             vdwioffsetptr0+vdwjidx0D,
1060                                             &c6_00,&c12_00);
1061
1062             /* EWALD ELECTROSTATICS */
1063
1064             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1065             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1066             ewitab           = _mm256_cvttpd_epi32(ewrt);
1067             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1068             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1069                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1070                                             &ewtabF,&ewtabFn);
1071             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1072             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1073
1074             /* LENNARD-JONES DISPERSION/REPULSION */
1075
1076             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1077             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1078
1079             fscal            = _mm256_add_pd(felec,fvdw);
1080
1081             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1082
1083             /* Calculate temporary vectorial force */
1084             tx               = _mm256_mul_pd(fscal,dx00);
1085             ty               = _mm256_mul_pd(fscal,dy00);
1086             tz               = _mm256_mul_pd(fscal,dz00);
1087
1088             /* Update vectorial force */
1089             fix0             = _mm256_add_pd(fix0,tx);
1090             fiy0             = _mm256_add_pd(fiy0,ty);
1091             fiz0             = _mm256_add_pd(fiz0,tz);
1092
1093             fjx0             = _mm256_add_pd(fjx0,tx);
1094             fjy0             = _mm256_add_pd(fjy0,ty);
1095             fjz0             = _mm256_add_pd(fjz0,tz);
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             r10              = _mm256_mul_pd(rsq10,rinv10);
1102             r10              = _mm256_andnot_pd(dummy_mask,r10);
1103
1104             /* Compute parameters for interactions between i and j atoms */
1105             qq10             = _mm256_mul_pd(iq1,jq0);
1106
1107             /* EWALD ELECTROSTATICS */
1108
1109             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1110             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1111             ewitab           = _mm256_cvttpd_epi32(ewrt);
1112             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1113             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1114                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1115                                             &ewtabF,&ewtabFn);
1116             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1117             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1118
1119             fscal            = felec;
1120
1121             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1122
1123             /* Calculate temporary vectorial force */
1124             tx               = _mm256_mul_pd(fscal,dx10);
1125             ty               = _mm256_mul_pd(fscal,dy10);
1126             tz               = _mm256_mul_pd(fscal,dz10);
1127
1128             /* Update vectorial force */
1129             fix1             = _mm256_add_pd(fix1,tx);
1130             fiy1             = _mm256_add_pd(fiy1,ty);
1131             fiz1             = _mm256_add_pd(fiz1,tz);
1132
1133             fjx0             = _mm256_add_pd(fjx0,tx);
1134             fjy0             = _mm256_add_pd(fjy0,ty);
1135             fjz0             = _mm256_add_pd(fjz0,tz);
1136
1137             /**************************
1138              * CALCULATE INTERACTIONS *
1139              **************************/
1140
1141             r20              = _mm256_mul_pd(rsq20,rinv20);
1142             r20              = _mm256_andnot_pd(dummy_mask,r20);
1143
1144             /* Compute parameters for interactions between i and j atoms */
1145             qq20             = _mm256_mul_pd(iq2,jq0);
1146
1147             /* EWALD ELECTROSTATICS */
1148
1149             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1150             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1151             ewitab           = _mm256_cvttpd_epi32(ewrt);
1152             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1153             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1154                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1155                                             &ewtabF,&ewtabFn);
1156             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1157             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1158
1159             fscal            = felec;
1160
1161             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1162
1163             /* Calculate temporary vectorial force */
1164             tx               = _mm256_mul_pd(fscal,dx20);
1165             ty               = _mm256_mul_pd(fscal,dy20);
1166             tz               = _mm256_mul_pd(fscal,dz20);
1167
1168             /* Update vectorial force */
1169             fix2             = _mm256_add_pd(fix2,tx);
1170             fiy2             = _mm256_add_pd(fiy2,ty);
1171             fiz2             = _mm256_add_pd(fiz2,tz);
1172
1173             fjx0             = _mm256_add_pd(fjx0,tx);
1174             fjy0             = _mm256_add_pd(fjy0,ty);
1175             fjz0             = _mm256_add_pd(fjz0,tz);
1176
1177             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1178             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1179             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1180             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1181
1182             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1183
1184             /* Inner loop uses 121 flops */
1185         }
1186
1187         /* End of innermost loop */
1188
1189         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1190                                                  f+i_coord_offset,fshift+i_shift_offset);
1191
1192         /* Increment number of inner iterations */
1193         inneriter                  += j_index_end - j_index_start;
1194
1195         /* Outer loop uses 18 flops */
1196     }
1197
1198     /* Increment number of outer iterations */
1199     outeriter        += nri;
1200
1201     /* Update outer/inner flops */
1202
1203     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*121);
1204 }