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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     int              nvdwtype;
  98     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 102     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 103     __m128i          ewitab;
 104     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 105     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 106     real             *ewtab;
 107     __m256d          dummy_mask,cutoff_mask;
 108     __m128           tmpmask0,tmpmask1;
 109     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 110     __m256d          one     = _mm256_set1_pd(1.0);
 111     __m256d          two     = _mm256_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm256_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 130     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 131     beta2            = _mm256_mul_pd(beta,beta);
 132     beta3            = _mm256_mul_pd(beta,beta2);
 133
 134     ewtab            = fr->ic->tabq_coul_FDV0;
 135     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 136     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[0];
 140     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 141     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 142     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 143     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm256_setzero_pd();
 179         fiy0             = _mm256_setzero_pd();
 180         fiz0             = _mm256_setzero_pd();
 181         fix1             = _mm256_setzero_pd();
 182         fiy1             = _mm256_setzero_pd();
 183         fiz1             = _mm256_setzero_pd();
 184         fix2             = _mm256_setzero_pd();
 185         fiy2             = _mm256_setzero_pd();
 186         fiz2             = _mm256_setzero_pd();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_pd();
 190         vvdwsum          = _mm256_setzero_pd();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 209                                                  &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm256_sub_pd(ix0,jx0);
 213             dy00             = _mm256_sub_pd(iy0,jy0);
 214             dz00             = _mm256_sub_pd(iz0,jz0);
 215             dx10             = _mm256_sub_pd(ix1,jx0);
 216             dy10             = _mm256_sub_pd(iy1,jy0);
 217             dz10             = _mm256_sub_pd(iz1,jz0);
 218             dx20             = _mm256_sub_pd(ix2,jx0);
 219             dy20             = _mm256_sub_pd(iy2,jy0);
 220             dz20             = _mm256_sub_pd(iz2,jz0);
 221
 222             /* Calculate squared distance and things based on it */
 223             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 224             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 225             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 226
 227             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 228             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 229             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 230
 231             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 232             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 233             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 234
 235             /* Load parameters for j particles */
 236             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 237                                                                  charge+jnrC+0,charge+jnrD+0);
 238             vdwjidx0A        = 2*vdwtype[jnrA+0];
 239             vdwjidx0B        = 2*vdwtype[jnrB+0];
 240             vdwjidx0C        = 2*vdwtype[jnrC+0];
 241             vdwjidx0D        = 2*vdwtype[jnrD+0];
 242
 243             fjx0             = _mm256_setzero_pd();
 244             fjy0             = _mm256_setzero_pd();
 245             fjz0             = _mm256_setzero_pd();
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             r00              = _mm256_mul_pd(rsq00,rinv00);
 252
 253             /* Compute parameters for interactions between i and j atoms */
 254             qq00             = _mm256_mul_pd(iq0,jq0);
 255             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 256                                             vdwioffsetptr0+vdwjidx0B,
 257                                             vdwioffsetptr0+vdwjidx0C,
 258                                             vdwioffsetptr0+vdwjidx0D,
 259                                             &c6_00,&c12_00);
 260
 261             /* EWALD ELECTROSTATICS */
 262
 263             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 264             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 265             ewitab           = _mm256_cvttpd_epi32(ewrt);
 266             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 267             ewitab           = _mm_slli_epi32(ewitab,2);
 268             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 269             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 270             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 271             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 272             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 273             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 274             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 275             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 276             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 277
 278             /* LENNARD-JONES DISPERSION/REPULSION */
 279
 280             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 281             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 282             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 283             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 284             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velecsum         = _mm256_add_pd(velecsum,velec);
 288             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 289
 290             fscal            = _mm256_add_pd(felec,fvdw);
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm256_mul_pd(fscal,dx00);
 294             ty               = _mm256_mul_pd(fscal,dy00);
 295             tz               = _mm256_mul_pd(fscal,dz00);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm256_add_pd(fix0,tx);
 299             fiy0             = _mm256_add_pd(fiy0,ty);
 300             fiz0             = _mm256_add_pd(fiz0,tz);
 301
 302             fjx0             = _mm256_add_pd(fjx0,tx);
 303             fjy0             = _mm256_add_pd(fjy0,ty);
 304             fjz0             = _mm256_add_pd(fjz0,tz);
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             r10              = _mm256_mul_pd(rsq10,rinv10);
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq10             = _mm256_mul_pd(iq1,jq0);
 314
 315             /* EWALD ELECTROSTATICS */
 316
 317             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 318             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 319             ewitab           = _mm256_cvttpd_epi32(ewrt);
 320             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 321             ewitab           = _mm_slli_epi32(ewitab,2);
 322             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 323             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 324             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 325             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 326             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 327             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 328             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 329             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 330             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velecsum         = _mm256_add_pd(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337             /* Calculate temporary vectorial force */
 338             tx               = _mm256_mul_pd(fscal,dx10);
 339             ty               = _mm256_mul_pd(fscal,dy10);
 340             tz               = _mm256_mul_pd(fscal,dz10);
 341
 342             /* Update vectorial force */
 343             fix1             = _mm256_add_pd(fix1,tx);
 344             fiy1             = _mm256_add_pd(fiy1,ty);
 345             fiz1             = _mm256_add_pd(fiz1,tz);
 346
 347             fjx0             = _mm256_add_pd(fjx0,tx);
 348             fjy0             = _mm256_add_pd(fjy0,ty);
 349             fjz0             = _mm256_add_pd(fjz0,tz);
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             r20              = _mm256_mul_pd(rsq20,rinv20);
 356
 357             /* Compute parameters for interactions between i and j atoms */
 358             qq20             = _mm256_mul_pd(iq2,jq0);
 359
 360             /* EWALD ELECTROSTATICS */
 361
 362             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 363             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 364             ewitab           = _mm256_cvttpd_epi32(ewrt);
 365             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 366             ewitab           = _mm_slli_epi32(ewitab,2);
 367             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 368             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 369             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 370             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 371             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 372             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 373             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 374             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 375             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velecsum         = _mm256_add_pd(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm256_mul_pd(fscal,dx20);
 384             ty               = _mm256_mul_pd(fscal,dy20);
 385             tz               = _mm256_mul_pd(fscal,dz20);
 386
 387             /* Update vectorial force */
 388             fix2             = _mm256_add_pd(fix2,tx);
 389             fiy2             = _mm256_add_pd(fiy2,ty);
 390             fiz2             = _mm256_add_pd(fiz2,tz);
 391
 392             fjx0             = _mm256_add_pd(fjx0,tx);
 393             fjy0             = _mm256_add_pd(fjy0,ty);
 394             fjz0             = _mm256_add_pd(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
 401             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 402
 403             /* Inner loop uses 138 flops */
 404         }
 405
 406         if(jidx<j_index_end)
 407         {
 408
 409             /* Get j neighbor index, and coordinate index */
 410             jnrlistA         = jjnr[jidx];
 411             jnrlistB         = jjnr[jidx+1];
 412             jnrlistC         = jjnr[jidx+2];
 413             jnrlistD         = jjnr[jidx+3];
 414             /* Sign of each element will be negative for non-real atoms.
 415              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 416              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 417              */
 418             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 419
 420             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 421             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 422             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 423
 424             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 425             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 426             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 427             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 428             j_coord_offsetA  = DIM*jnrA;
 429             j_coord_offsetB  = DIM*jnrB;
 430             j_coord_offsetC  = DIM*jnrC;
 431             j_coord_offsetD  = DIM*jnrD;
 432
 433             /* load j atom coordinates */
 434             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 435                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 436                                                  &jx0,&jy0,&jz0);
 437
 438             /* Calculate displacement vector */
 439             dx00             = _mm256_sub_pd(ix0,jx0);
 440             dy00             = _mm256_sub_pd(iy0,jy0);
 441             dz00             = _mm256_sub_pd(iz0,jz0);
 442             dx10             = _mm256_sub_pd(ix1,jx0);
 443             dy10             = _mm256_sub_pd(iy1,jy0);
 444             dz10             = _mm256_sub_pd(iz1,jz0);
 445             dx20             = _mm256_sub_pd(ix2,jx0);
 446             dy20             = _mm256_sub_pd(iy2,jy0);
 447             dz20             = _mm256_sub_pd(iz2,jz0);
 448
 449             /* Calculate squared distance and things based on it */
 450             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 451             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 452             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 453
 454             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 455             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 456             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 457
 458             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 459             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 460             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 461
 462             /* Load parameters for j particles */
 463             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 464                                                                  charge+jnrC+0,charge+jnrD+0);
 465             vdwjidx0A        = 2*vdwtype[jnrA+0];
 466             vdwjidx0B        = 2*vdwtype[jnrB+0];
 467             vdwjidx0C        = 2*vdwtype[jnrC+0];
 468             vdwjidx0D        = 2*vdwtype[jnrD+0];
 469
 470             fjx0             = _mm256_setzero_pd();
 471             fjy0             = _mm256_setzero_pd();
 472             fjz0             = _mm256_setzero_pd();
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             r00              = _mm256_mul_pd(rsq00,rinv00);
 479             r00              = _mm256_andnot_pd(dummy_mask,r00);
 480
 481             /* Compute parameters for interactions between i and j atoms */
 482             qq00             = _mm256_mul_pd(iq0,jq0);
 483             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 484                                             vdwioffsetptr0+vdwjidx0B,
 485                                             vdwioffsetptr0+vdwjidx0C,
 486                                             vdwioffsetptr0+vdwjidx0D,
 487                                             &c6_00,&c12_00);
 488
 489             /* EWALD ELECTROSTATICS */
 490
 491             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 492             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 493             ewitab           = _mm256_cvttpd_epi32(ewrt);
 494             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 495             ewitab           = _mm_slli_epi32(ewitab,2);
 496             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 497             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 498             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 499             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 500             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 501             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 502             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 503             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 504             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 505
 506             /* LENNARD-JONES DISPERSION/REPULSION */
 507
 508             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 509             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 510             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 511             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 512             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 513
 514             /* Update potential sum for this i atom from the interaction with this j atom. */
 515             velec            = _mm256_andnot_pd(dummy_mask,velec);
 516             velecsum         = _mm256_add_pd(velecsum,velec);
 517             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 518             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 519
 520             fscal            = _mm256_add_pd(felec,fvdw);
 521
 522             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm256_mul_pd(fscal,dx00);
 526             ty               = _mm256_mul_pd(fscal,dy00);
 527             tz               = _mm256_mul_pd(fscal,dz00);
 528
 529             /* Update vectorial force */
 530             fix0             = _mm256_add_pd(fix0,tx);
 531             fiy0             = _mm256_add_pd(fiy0,ty);
 532             fiz0             = _mm256_add_pd(fiz0,tz);
 533
 534             fjx0             = _mm256_add_pd(fjx0,tx);
 535             fjy0             = _mm256_add_pd(fjy0,ty);
 536             fjz0             = _mm256_add_pd(fjz0,tz);
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             r10              = _mm256_mul_pd(rsq10,rinv10);
 543             r10              = _mm256_andnot_pd(dummy_mask,r10);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq10             = _mm256_mul_pd(iq1,jq0);
 547
 548             /* EWALD ELECTROSTATICS */
 549
 550             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 551             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 552             ewitab           = _mm256_cvttpd_epi32(ewrt);
 553             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 554             ewitab           = _mm_slli_epi32(ewitab,2);
 555             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 556             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 557             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 558             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 559             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 560             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 561             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 562             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 563             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             velec            = _mm256_andnot_pd(dummy_mask,velec);
 567             velecsum         = _mm256_add_pd(velecsum,velec);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm256_mul_pd(fscal,dx10);
 575             ty               = _mm256_mul_pd(fscal,dy10);
 576             tz               = _mm256_mul_pd(fscal,dz10);
 577
 578             /* Update vectorial force */
 579             fix1             = _mm256_add_pd(fix1,tx);
 580             fiy1             = _mm256_add_pd(fiy1,ty);
 581             fiz1             = _mm256_add_pd(fiz1,tz);
 582
 583             fjx0             = _mm256_add_pd(fjx0,tx);
 584             fjy0             = _mm256_add_pd(fjy0,ty);
 585             fjz0             = _mm256_add_pd(fjz0,tz);
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             r20              = _mm256_mul_pd(rsq20,rinv20);
 592             r20              = _mm256_andnot_pd(dummy_mask,r20);
 593
 594             /* Compute parameters for interactions between i and j atoms */
 595             qq20             = _mm256_mul_pd(iq2,jq0);
 596
 597             /* EWALD ELECTROSTATICS */
 598
 599             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 600             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 601             ewitab           = _mm256_cvttpd_epi32(ewrt);
 602             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 603             ewitab           = _mm_slli_epi32(ewitab,2);
 604             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 605             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 606             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 607             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 608             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 609             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 610             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 611             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 612             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velec            = _mm256_andnot_pd(dummy_mask,velec);
 616             velecsum         = _mm256_add_pd(velecsum,velec);
 617
 618             fscal            = felec;
 619
 620             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 621
 622             /* Calculate temporary vectorial force */
 623             tx               = _mm256_mul_pd(fscal,dx20);
 624             ty               = _mm256_mul_pd(fscal,dy20);
 625             tz               = _mm256_mul_pd(fscal,dz20);
 626
 627             /* Update vectorial force */
 628             fix2             = _mm256_add_pd(fix2,tx);
 629             fiy2             = _mm256_add_pd(fiy2,ty);
 630             fiz2             = _mm256_add_pd(fiz2,tz);
 631
 632             fjx0             = _mm256_add_pd(fjx0,tx);
 633             fjy0             = _mm256_add_pd(fjy0,ty);
 634             fjz0             = _mm256_add_pd(fjz0,tz);
 635
 636             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 637             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 638             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 639             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 640
 641             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 642
 643             /* Inner loop uses 141 flops */
 644         }
 645
 646         /* End of innermost loop */
 647
 648         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 649                                                  f+i_coord_offset,fshift+i_shift_offset);
 650
 651         ggid                        = gid[iidx];
 652         /* Update potential energies */
 653         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 654         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 655
 656         /* Increment number of inner iterations */
 657         inneriter                  += j_index_end - j_index_start;
 658
 659         /* Outer loop uses 20 flops */
 660     }
 661
 662     /* Increment number of outer iterations */
 663     outeriter        += nri;
 664
 665     /* Update outer/inner flops */
 666
 667     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*141);
 668 }
 669 /*
 670  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_double
 671  * Electrostatics interaction: Ewald
 672  * VdW interaction:            LennardJones
 673  * Geometry:                   Water3-Particle
 674  * Calculate force/pot:        Force
 675  */
 676 void
 677 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_double
 678                     (t_nblist                    * gmx_restrict       nlist,
 679                      rvec                        * gmx_restrict          xx,
 680                      rvec                        * gmx_restrict          ff,
 681                      t_forcerec                  * gmx_restrict          fr,
 682                      t_mdatoms                   * gmx_restrict     mdatoms,
 683                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 684                      t_nrnb                      * gmx_restrict        nrnb)
 685 {
 686     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 687      * just 0 for non-waters.
 688      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 689      * jnr indices corresponding to data put in the four positions in the SIMD register.
 690      */
 691     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 692     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 693     int              jnrA,jnrB,jnrC,jnrD;
 694     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 695     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 696     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 697     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 698     real             rcutoff_scalar;
 699     real             *shiftvec,*fshift,*x,*f;
 700     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 701     real             scratch[4*DIM];
 702     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 703     real *           vdwioffsetptr0;
 704     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 705     real *           vdwioffsetptr1;
 706     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 707     real *           vdwioffsetptr2;
 708     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 709     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 710     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 711     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 712     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 713     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 714     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 715     real             *charge;
 716     int              nvdwtype;
 717     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 718     int              *vdwtype;
 719     real             *vdwparam;
 720     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 721     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 722     __m128i          ewitab;
 723     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 724     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 725     real             *ewtab;
 726     __m256d          dummy_mask,cutoff_mask;
 727     __m128           tmpmask0,tmpmask1;
 728     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 729     __m256d          one     = _mm256_set1_pd(1.0);
 730     __m256d          two     = _mm256_set1_pd(2.0);
 731     x                = xx[0];
 732     f                = ff[0];
 733
 734     nri              = nlist->nri;
 735     iinr             = nlist->iinr;
 736     jindex           = nlist->jindex;
 737     jjnr             = nlist->jjnr;
 738     shiftidx         = nlist->shift;
 739     gid              = nlist->gid;
 740     shiftvec         = fr->shift_vec[0];
 741     fshift           = fr->fshift[0];
 742     facel            = _mm256_set1_pd(fr->epsfac);
 743     charge           = mdatoms->chargeA;
 744     nvdwtype         = fr->ntype;
 745     vdwparam         = fr->nbfp;
 746     vdwtype          = mdatoms->typeA;
 747
 748     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 749     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 750     beta2            = _mm256_mul_pd(beta,beta);
 751     beta3            = _mm256_mul_pd(beta,beta2);
 752
 753     ewtab            = fr->ic->tabq_coul_F;
 754     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 755     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 756
 757     /* Setup water-specific parameters */
 758     inr              = nlist->iinr[0];
 759     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 760     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 761     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 762     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 763
 764     /* Avoid stupid compiler warnings */
 765     jnrA = jnrB = jnrC = jnrD = 0;
 766     j_coord_offsetA = 0;
 767     j_coord_offsetB = 0;
 768     j_coord_offsetC = 0;
 769     j_coord_offsetD = 0;
 770
 771     outeriter        = 0;
 772     inneriter        = 0;
 773
 774     for(iidx=0;iidx<4*DIM;iidx++)
 775     {
 776         scratch[iidx] = 0.0;
 777     }
 778
 779     /* Start outer loop over neighborlists */
 780     for(iidx=0; iidx<nri; iidx++)
 781     {
 782         /* Load shift vector for this list */
 783         i_shift_offset   = DIM*shiftidx[iidx];
 784
 785         /* Load limits for loop over neighbors */
 786         j_index_start    = jindex[iidx];
 787         j_index_end      = jindex[iidx+1];
 788
 789         /* Get outer coordinate index */
 790         inr              = iinr[iidx];
 791         i_coord_offset   = DIM*inr;
 792
 793         /* Load i particle coords and add shift vector */
 794         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 795                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 796
 797         fix0             = _mm256_setzero_pd();
 798         fiy0             = _mm256_setzero_pd();
 799         fiz0             = _mm256_setzero_pd();
 800         fix1             = _mm256_setzero_pd();
 801         fiy1             = _mm256_setzero_pd();
 802         fiz1             = _mm256_setzero_pd();
 803         fix2             = _mm256_setzero_pd();
 804         fiy2             = _mm256_setzero_pd();
 805         fiz2             = _mm256_setzero_pd();
 806
 807         /* Start inner kernel loop */
 808         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 809         {
 810
 811             /* Get j neighbor index, and coordinate index */
 812             jnrA             = jjnr[jidx];
 813             jnrB             = jjnr[jidx+1];
 814             jnrC             = jjnr[jidx+2];
 815             jnrD             = jjnr[jidx+3];
 816             j_coord_offsetA  = DIM*jnrA;
 817             j_coord_offsetB  = DIM*jnrB;
 818             j_coord_offsetC  = DIM*jnrC;
 819             j_coord_offsetD  = DIM*jnrD;
 820
 821             /* load j atom coordinates */
 822             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 823                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 824                                                  &jx0,&jy0,&jz0);
 825
 826             /* Calculate displacement vector */
 827             dx00             = _mm256_sub_pd(ix0,jx0);
 828             dy00             = _mm256_sub_pd(iy0,jy0);
 829             dz00             = _mm256_sub_pd(iz0,jz0);
 830             dx10             = _mm256_sub_pd(ix1,jx0);
 831             dy10             = _mm256_sub_pd(iy1,jy0);
 832             dz10             = _mm256_sub_pd(iz1,jz0);
 833             dx20             = _mm256_sub_pd(ix2,jx0);
 834             dy20             = _mm256_sub_pd(iy2,jy0);
 835             dz20             = _mm256_sub_pd(iz2,jz0);
 836
 837             /* Calculate squared distance and things based on it */
 838             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 839             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 840             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 841
 842             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 843             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 844             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 845
 846             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 847             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 848             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 849
 850             /* Load parameters for j particles */
 851             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 852                                                                  charge+jnrC+0,charge+jnrD+0);
 853             vdwjidx0A        = 2*vdwtype[jnrA+0];
 854             vdwjidx0B        = 2*vdwtype[jnrB+0];
 855             vdwjidx0C        = 2*vdwtype[jnrC+0];
 856             vdwjidx0D        = 2*vdwtype[jnrD+0];
 857
 858             fjx0             = _mm256_setzero_pd();
 859             fjy0             = _mm256_setzero_pd();
 860             fjz0             = _mm256_setzero_pd();
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             r00              = _mm256_mul_pd(rsq00,rinv00);
 867
 868             /* Compute parameters for interactions between i and j atoms */
 869             qq00             = _mm256_mul_pd(iq0,jq0);
 870             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 871                                             vdwioffsetptr0+vdwjidx0B,
 872                                             vdwioffsetptr0+vdwjidx0C,
 873                                             vdwioffsetptr0+vdwjidx0D,
 874                                             &c6_00,&c12_00);
 875
 876             /* EWALD ELECTROSTATICS */
 877
 878             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 879             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 880             ewitab           = _mm256_cvttpd_epi32(ewrt);
 881             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 882             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 883                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 884                                             &ewtabF,&ewtabFn);
 885             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 886             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 887
 888             /* LENNARD-JONES DISPERSION/REPULSION */
 889
 890             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 891             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 892
 893             fscal            = _mm256_add_pd(felec,fvdw);
 894
 895             /* Calculate temporary vectorial force */
 896             tx               = _mm256_mul_pd(fscal,dx00);
 897             ty               = _mm256_mul_pd(fscal,dy00);
 898             tz               = _mm256_mul_pd(fscal,dz00);
 899
 900             /* Update vectorial force */
 901             fix0             = _mm256_add_pd(fix0,tx);
 902             fiy0             = _mm256_add_pd(fiy0,ty);
 903             fiz0             = _mm256_add_pd(fiz0,tz);
 904
 905             fjx0             = _mm256_add_pd(fjx0,tx);
 906             fjy0             = _mm256_add_pd(fjy0,ty);
 907             fjz0             = _mm256_add_pd(fjz0,tz);
 908
 909             /**************************
 910              * CALCULATE INTERACTIONS *
 911              **************************/
 912
 913             r10              = _mm256_mul_pd(rsq10,rinv10);
 914
 915             /* Compute parameters for interactions between i and j atoms */
 916             qq10             = _mm256_mul_pd(iq1,jq0);
 917
 918             /* EWALD ELECTROSTATICS */
 919
 920             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 921             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 922             ewitab           = _mm256_cvttpd_epi32(ewrt);
 923             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 924             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 925                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 926                                             &ewtabF,&ewtabFn);
 927             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 928             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 929
 930             fscal            = felec;
 931
 932             /* Calculate temporary vectorial force */
 933             tx               = _mm256_mul_pd(fscal,dx10);
 934             ty               = _mm256_mul_pd(fscal,dy10);
 935             tz               = _mm256_mul_pd(fscal,dz10);
 936
 937             /* Update vectorial force */
 938             fix1             = _mm256_add_pd(fix1,tx);
 939             fiy1             = _mm256_add_pd(fiy1,ty);
 940             fiz1             = _mm256_add_pd(fiz1,tz);
 941
 942             fjx0             = _mm256_add_pd(fjx0,tx);
 943             fjy0             = _mm256_add_pd(fjy0,ty);
 944             fjz0             = _mm256_add_pd(fjz0,tz);
 945
 946             /**************************
 947              * CALCULATE INTERACTIONS *
 948              **************************/
 949
 950             r20              = _mm256_mul_pd(rsq20,rinv20);
 951
 952             /* Compute parameters for interactions between i and j atoms */
 953             qq20             = _mm256_mul_pd(iq2,jq0);
 954
 955             /* EWALD ELECTROSTATICS */
 956
 957             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 958             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 959             ewitab           = _mm256_cvttpd_epi32(ewrt);
 960             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 961             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 962                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 963                                             &ewtabF,&ewtabFn);
 964             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 965             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 966
 967             fscal            = felec;
 968
 969             /* Calculate temporary vectorial force */
 970             tx               = _mm256_mul_pd(fscal,dx20);
 971             ty               = _mm256_mul_pd(fscal,dy20);
 972             tz               = _mm256_mul_pd(fscal,dz20);
 973
 974             /* Update vectorial force */
 975             fix2             = _mm256_add_pd(fix2,tx);
 976             fiy2             = _mm256_add_pd(fiy2,ty);
 977             fiz2             = _mm256_add_pd(fiz2,tz);
 978
 979             fjx0             = _mm256_add_pd(fjx0,tx);
 980             fjy0             = _mm256_add_pd(fjy0,ty);
 981             fjz0             = _mm256_add_pd(fjz0,tz);
 982
 983             fjptrA             = f+j_coord_offsetA;
 984             fjptrB             = f+j_coord_offsetB;
 985             fjptrC             = f+j_coord_offsetC;
 986             fjptrD             = f+j_coord_offsetD;
 987
 988             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 989
 990             /* Inner loop uses 118 flops */
 991         }
 992
 993         if(jidx<j_index_end)
 994         {
 995
 996             /* Get j neighbor index, and coordinate index */
 997             jnrlistA         = jjnr[jidx];
 998             jnrlistB         = jjnr[jidx+1];
 999             jnrlistC         = jjnr[jidx+2];
1000             jnrlistD         = jjnr[jidx+3];
1001             /* Sign of each element will be negative for non-real atoms.
1002              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1003              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1004              */
1005             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1006
1007             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1008             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1009             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1010
1011             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1012             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1013             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1014             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1015             j_coord_offsetA  = DIM*jnrA;
1016             j_coord_offsetB  = DIM*jnrB;
1017             j_coord_offsetC  = DIM*jnrC;
1018             j_coord_offsetD  = DIM*jnrD;
1019
1020             /* load j atom coordinates */
1021             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1022                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1023                                                  &jx0,&jy0,&jz0);
1024
1025             /* Calculate displacement vector */
1026             dx00             = _mm256_sub_pd(ix0,jx0);
1027             dy00             = _mm256_sub_pd(iy0,jy0);
1028             dz00             = _mm256_sub_pd(iz0,jz0);
1029             dx10             = _mm256_sub_pd(ix1,jx0);
1030             dy10             = _mm256_sub_pd(iy1,jy0);
1031             dz10             = _mm256_sub_pd(iz1,jz0);
1032             dx20             = _mm256_sub_pd(ix2,jx0);
1033             dy20             = _mm256_sub_pd(iy2,jy0);
1034             dz20             = _mm256_sub_pd(iz2,jz0);
1035
1036             /* Calculate squared distance and things based on it */
1037             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1038             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1039             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1040
1041             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1042             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1043             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1044
1045             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1046             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1047             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1048
1049             /* Load parameters for j particles */
1050             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1051                                                                  charge+jnrC+0,charge+jnrD+0);
1052             vdwjidx0A        = 2*vdwtype[jnrA+0];
1053             vdwjidx0B        = 2*vdwtype[jnrB+0];
1054             vdwjidx0C        = 2*vdwtype[jnrC+0];
1055             vdwjidx0D        = 2*vdwtype[jnrD+0];
1056
1057             fjx0             = _mm256_setzero_pd();
1058             fjy0             = _mm256_setzero_pd();
1059             fjz0             = _mm256_setzero_pd();
1060
1061             /**************************
1062              * CALCULATE INTERACTIONS *
1063              **************************/
1064
1065             r00              = _mm256_mul_pd(rsq00,rinv00);
1066             r00              = _mm256_andnot_pd(dummy_mask,r00);
1067
1068             /* Compute parameters for interactions between i and j atoms */
1069             qq00             = _mm256_mul_pd(iq0,jq0);
1070             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1071                                             vdwioffsetptr0+vdwjidx0B,
1072                                             vdwioffsetptr0+vdwjidx0C,
1073                                             vdwioffsetptr0+vdwjidx0D,
1074                                             &c6_00,&c12_00);
1075
1076             /* EWALD ELECTROSTATICS */
1077
1078             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1079             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1080             ewitab           = _mm256_cvttpd_epi32(ewrt);
1081             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1082             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1083                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1084                                             &ewtabF,&ewtabFn);
1085             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1086             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1087
1088             /* LENNARD-JONES DISPERSION/REPULSION */
1089
1090             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1091             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1092
1093             fscal            = _mm256_add_pd(felec,fvdw);
1094
1095             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1096
1097             /* Calculate temporary vectorial force */
1098             tx               = _mm256_mul_pd(fscal,dx00);
1099             ty               = _mm256_mul_pd(fscal,dy00);
1100             tz               = _mm256_mul_pd(fscal,dz00);
1101
1102             /* Update vectorial force */
1103             fix0             = _mm256_add_pd(fix0,tx);
1104             fiy0             = _mm256_add_pd(fiy0,ty);
1105             fiz0             = _mm256_add_pd(fiz0,tz);
1106
1107             fjx0             = _mm256_add_pd(fjx0,tx);
1108             fjy0             = _mm256_add_pd(fjy0,ty);
1109             fjz0             = _mm256_add_pd(fjz0,tz);
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             r10              = _mm256_mul_pd(rsq10,rinv10);
1116             r10              = _mm256_andnot_pd(dummy_mask,r10);
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq10             = _mm256_mul_pd(iq1,jq0);
1120
1121             /* EWALD ELECTROSTATICS */
1122
1123             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1124             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1125             ewitab           = _mm256_cvttpd_epi32(ewrt);
1126             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1127             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1128                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1129                                             &ewtabF,&ewtabFn);
1130             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1131             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1132
1133             fscal            = felec;
1134
1135             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1136
1137             /* Calculate temporary vectorial force */
1138             tx               = _mm256_mul_pd(fscal,dx10);
1139             ty               = _mm256_mul_pd(fscal,dy10);
1140             tz               = _mm256_mul_pd(fscal,dz10);
1141
1142             /* Update vectorial force */
1143             fix1             = _mm256_add_pd(fix1,tx);
1144             fiy1             = _mm256_add_pd(fiy1,ty);
1145             fiz1             = _mm256_add_pd(fiz1,tz);
1146
1147             fjx0             = _mm256_add_pd(fjx0,tx);
1148             fjy0             = _mm256_add_pd(fjy0,ty);
1149             fjz0             = _mm256_add_pd(fjz0,tz);
1150
1151             /**************************
1152              * CALCULATE INTERACTIONS *
1153              **************************/
1154
1155             r20              = _mm256_mul_pd(rsq20,rinv20);
1156             r20              = _mm256_andnot_pd(dummy_mask,r20);
1157
1158             /* Compute parameters for interactions between i and j atoms */
1159             qq20             = _mm256_mul_pd(iq2,jq0);
1160
1161             /* EWALD ELECTROSTATICS */
1162
1163             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1164             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1165             ewitab           = _mm256_cvttpd_epi32(ewrt);
1166             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1167             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1168                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1169                                             &ewtabF,&ewtabFn);
1170             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1171             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1172
1173             fscal            = felec;
1174
1175             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1176
1177             /* Calculate temporary vectorial force */
1178             tx               = _mm256_mul_pd(fscal,dx20);
1179             ty               = _mm256_mul_pd(fscal,dy20);
1180             tz               = _mm256_mul_pd(fscal,dz20);
1181
1182             /* Update vectorial force */
1183             fix2             = _mm256_add_pd(fix2,tx);
1184             fiy2             = _mm256_add_pd(fiy2,ty);
1185             fiz2             = _mm256_add_pd(fiz2,tz);
1186
1187             fjx0             = _mm256_add_pd(fjx0,tx);
1188             fjy0             = _mm256_add_pd(fjy0,ty);
1189             fjz0             = _mm256_add_pd(fjz0,tz);
1190
1191             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1192             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1193             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1194             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1195
1196             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1197
1198             /* Inner loop uses 121 flops */
1199         }
1200
1201         /* End of innermost loop */
1202
1203         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1204                                                  f+i_coord_offset,fshift+i_shift_offset);
1205
1206         /* Increment number of inner iterations */
1207         inneriter                  += j_index_end - j_index_start;
1208
1209         /* Outer loop uses 18 flops */
1210     }
1211
1212     /* Increment number of outer iterations */
1213     outeriter        += nri;
1214
1215     /* Update outer/inner flops */
1216
1217     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*121);
1218 }