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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __m128i          ewitab;
 106     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 108     real             *ewtab;
 109     __m256d          dummy_mask,cutoff_mask;
 110     __m128           tmpmask0,tmpmask1;
 111     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 112     __m256d          one     = _mm256_set1_pd(1.0);
 113     __m256d          two     = _mm256_set1_pd(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_pd(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 132     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 133     beta2            = _mm256_mul_pd(beta,beta);
 134     beta3            = _mm256_mul_pd(beta,beta2);
 135
 136     ewtab            = fr->ic->tabq_coul_FDV0;
 137     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 138     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 143     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 144     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 145     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm256_setzero_pd();
 181         fiy0             = _mm256_setzero_pd();
 182         fiz0             = _mm256_setzero_pd();
 183         fix1             = _mm256_setzero_pd();
 184         fiy1             = _mm256_setzero_pd();
 185         fiz1             = _mm256_setzero_pd();
 186         fix2             = _mm256_setzero_pd();
 187         fiy2             = _mm256_setzero_pd();
 188         fiz2             = _mm256_setzero_pd();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm256_setzero_pd();
 192         vvdwsum          = _mm256_setzero_pd();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_pd(ix0,jx0);
 215             dy00             = _mm256_sub_pd(iy0,jy0);
 216             dz00             = _mm256_sub_pd(iz0,jz0);
 217             dx10             = _mm256_sub_pd(ix1,jx0);
 218             dy10             = _mm256_sub_pd(iy1,jy0);
 219             dz10             = _mm256_sub_pd(iz1,jz0);
 220             dx20             = _mm256_sub_pd(ix2,jx0);
 221             dy20             = _mm256_sub_pd(iy2,jy0);
 222             dz20             = _mm256_sub_pd(iz2,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 226             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 227             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 228
 229             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 230             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 231             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 232
 233             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 234             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 235             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 239                                                                  charge+jnrC+0,charge+jnrD+0);
 240             vdwjidx0A        = 2*vdwtype[jnrA+0];
 241             vdwjidx0B        = 2*vdwtype[jnrB+0];
 242             vdwjidx0C        = 2*vdwtype[jnrC+0];
 243             vdwjidx0D        = 2*vdwtype[jnrD+0];
 244
 245             fjx0             = _mm256_setzero_pd();
 246             fjy0             = _mm256_setzero_pd();
 247             fjz0             = _mm256_setzero_pd();
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             r00              = _mm256_mul_pd(rsq00,rinv00);
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             qq00             = _mm256_mul_pd(iq0,jq0);
 257             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 258                                             vdwioffsetptr0+vdwjidx0B,
 259                                             vdwioffsetptr0+vdwjidx0C,
 260                                             vdwioffsetptr0+vdwjidx0D,
 261                                             &c6_00,&c12_00);
 262
 263             /* EWALD ELECTROSTATICS */
 264
 265             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 266             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 267             ewitab           = _mm256_cvttpd_epi32(ewrt);
 268             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 269             ewitab           = _mm_slli_epi32(ewitab,2);
 270             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 271             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 272             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 273             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 274             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 275             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 276             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 277             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 278             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 279
 280             /* LENNARD-JONES DISPERSION/REPULSION */
 281
 282             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 283             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 284             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 285             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 286             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             velecsum         = _mm256_add_pd(velecsum,velec);
 290             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 291
 292             fscal            = _mm256_add_pd(felec,fvdw);
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm256_mul_pd(fscal,dx00);
 296             ty               = _mm256_mul_pd(fscal,dy00);
 297             tz               = _mm256_mul_pd(fscal,dz00);
 298
 299             /* Update vectorial force */
 300             fix0             = _mm256_add_pd(fix0,tx);
 301             fiy0             = _mm256_add_pd(fiy0,ty);
 302             fiz0             = _mm256_add_pd(fiz0,tz);
 303
 304             fjx0             = _mm256_add_pd(fjx0,tx);
 305             fjy0             = _mm256_add_pd(fjy0,ty);
 306             fjz0             = _mm256_add_pd(fjz0,tz);
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             r10              = _mm256_mul_pd(rsq10,rinv10);
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq10             = _mm256_mul_pd(iq1,jq0);
 316
 317             /* EWALD ELECTROSTATICS */
 318
 319             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 320             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 321             ewitab           = _mm256_cvttpd_epi32(ewrt);
 322             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 323             ewitab           = _mm_slli_epi32(ewitab,2);
 324             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 325             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 326             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 327             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 328             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 329             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 330             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 331             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 332             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velecsum         = _mm256_add_pd(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             /* Calculate temporary vectorial force */
 340             tx               = _mm256_mul_pd(fscal,dx10);
 341             ty               = _mm256_mul_pd(fscal,dy10);
 342             tz               = _mm256_mul_pd(fscal,dz10);
 343
 344             /* Update vectorial force */
 345             fix1             = _mm256_add_pd(fix1,tx);
 346             fiy1             = _mm256_add_pd(fiy1,ty);
 347             fiz1             = _mm256_add_pd(fiz1,tz);
 348
 349             fjx0             = _mm256_add_pd(fjx0,tx);
 350             fjy0             = _mm256_add_pd(fjy0,ty);
 351             fjz0             = _mm256_add_pd(fjz0,tz);
 352
 353             /**************************
 354              * CALCULATE INTERACTIONS *
 355              **************************/
 356
 357             r20              = _mm256_mul_pd(rsq20,rinv20);
 358
 359             /* Compute parameters for interactions between i and j atoms */
 360             qq20             = _mm256_mul_pd(iq2,jq0);
 361
 362             /* EWALD ELECTROSTATICS */
 363
 364             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 365             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 366             ewitab           = _mm256_cvttpd_epi32(ewrt);
 367             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 368             ewitab           = _mm_slli_epi32(ewitab,2);
 369             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 370             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 371             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 372             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 373             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 374             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 375             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 376             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 377             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm256_add_pd(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384             /* Calculate temporary vectorial force */
 385             tx               = _mm256_mul_pd(fscal,dx20);
 386             ty               = _mm256_mul_pd(fscal,dy20);
 387             tz               = _mm256_mul_pd(fscal,dz20);
 388
 389             /* Update vectorial force */
 390             fix2             = _mm256_add_pd(fix2,tx);
 391             fiy2             = _mm256_add_pd(fiy2,ty);
 392             fiz2             = _mm256_add_pd(fiz2,tz);
 393
 394             fjx0             = _mm256_add_pd(fjx0,tx);
 395             fjy0             = _mm256_add_pd(fjy0,ty);
 396             fjz0             = _mm256_add_pd(fjz0,tz);
 397
 398             fjptrA             = f+j_coord_offsetA;
 399             fjptrB             = f+j_coord_offsetB;
 400             fjptrC             = f+j_coord_offsetC;
 401             fjptrD             = f+j_coord_offsetD;
 402
 403             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 404
 405             /* Inner loop uses 138 flops */
 406         }
 407
 408         if(jidx<j_index_end)
 409         {
 410
 411             /* Get j neighbor index, and coordinate index */
 412             jnrlistA         = jjnr[jidx];
 413             jnrlistB         = jjnr[jidx+1];
 414             jnrlistC         = jjnr[jidx+2];
 415             jnrlistD         = jjnr[jidx+3];
 416             /* Sign of each element will be negative for non-real atoms.
 417              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 418              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 419              */
 420             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 421
 422             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 423             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 424             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 425
 426             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 427             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 428             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 429             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 430             j_coord_offsetA  = DIM*jnrA;
 431             j_coord_offsetB  = DIM*jnrB;
 432             j_coord_offsetC  = DIM*jnrC;
 433             j_coord_offsetD  = DIM*jnrD;
 434
 435             /* load j atom coordinates */
 436             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 437                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 438                                                  &jx0,&jy0,&jz0);
 439
 440             /* Calculate displacement vector */
 441             dx00             = _mm256_sub_pd(ix0,jx0);
 442             dy00             = _mm256_sub_pd(iy0,jy0);
 443             dz00             = _mm256_sub_pd(iz0,jz0);
 444             dx10             = _mm256_sub_pd(ix1,jx0);
 445             dy10             = _mm256_sub_pd(iy1,jy0);
 446             dz10             = _mm256_sub_pd(iz1,jz0);
 447             dx20             = _mm256_sub_pd(ix2,jx0);
 448             dy20             = _mm256_sub_pd(iy2,jy0);
 449             dz20             = _mm256_sub_pd(iz2,jz0);
 450
 451             /* Calculate squared distance and things based on it */
 452             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 453             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 454             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 455
 456             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 457             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 458             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 459
 460             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 461             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 462             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 463
 464             /* Load parameters for j particles */
 465             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 466                                                                  charge+jnrC+0,charge+jnrD+0);
 467             vdwjidx0A        = 2*vdwtype[jnrA+0];
 468             vdwjidx0B        = 2*vdwtype[jnrB+0];
 469             vdwjidx0C        = 2*vdwtype[jnrC+0];
 470             vdwjidx0D        = 2*vdwtype[jnrD+0];
 471
 472             fjx0             = _mm256_setzero_pd();
 473             fjy0             = _mm256_setzero_pd();
 474             fjz0             = _mm256_setzero_pd();
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             r00              = _mm256_mul_pd(rsq00,rinv00);
 481             r00              = _mm256_andnot_pd(dummy_mask,r00);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq00             = _mm256_mul_pd(iq0,jq0);
 485             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 486                                             vdwioffsetptr0+vdwjidx0B,
 487                                             vdwioffsetptr0+vdwjidx0C,
 488                                             vdwioffsetptr0+vdwjidx0D,
 489                                             &c6_00,&c12_00);
 490
 491             /* EWALD ELECTROSTATICS */
 492
 493             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 494             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 495             ewitab           = _mm256_cvttpd_epi32(ewrt);
 496             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 497             ewitab           = _mm_slli_epi32(ewitab,2);
 498             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 499             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 500             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 501             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 502             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 503             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 504             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 505             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 506             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 507
 508             /* LENNARD-JONES DISPERSION/REPULSION */
 509
 510             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 511             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 512             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 513             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 514             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm256_andnot_pd(dummy_mask,velec);
 518             velecsum         = _mm256_add_pd(velecsum,velec);
 519             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 520             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 521
 522             fscal            = _mm256_add_pd(felec,fvdw);
 523
 524             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 525
 526             /* Calculate temporary vectorial force */
 527             tx               = _mm256_mul_pd(fscal,dx00);
 528             ty               = _mm256_mul_pd(fscal,dy00);
 529             tz               = _mm256_mul_pd(fscal,dz00);
 530
 531             /* Update vectorial force */
 532             fix0             = _mm256_add_pd(fix0,tx);
 533             fiy0             = _mm256_add_pd(fiy0,ty);
 534             fiz0             = _mm256_add_pd(fiz0,tz);
 535
 536             fjx0             = _mm256_add_pd(fjx0,tx);
 537             fjy0             = _mm256_add_pd(fjy0,ty);
 538             fjz0             = _mm256_add_pd(fjz0,tz);
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             r10              = _mm256_mul_pd(rsq10,rinv10);
 545             r10              = _mm256_andnot_pd(dummy_mask,r10);
 546
 547             /* Compute parameters for interactions between i and j atoms */
 548             qq10             = _mm256_mul_pd(iq1,jq0);
 549
 550             /* EWALD ELECTROSTATICS */
 551
 552             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 553             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 554             ewitab           = _mm256_cvttpd_epi32(ewrt);
 555             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 556             ewitab           = _mm_slli_epi32(ewitab,2);
 557             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 558             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 559             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 560             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 561             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 562             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 563             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 564             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 565             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velec            = _mm256_andnot_pd(dummy_mask,velec);
 569             velecsum         = _mm256_add_pd(velecsum,velec);
 570
 571             fscal            = felec;
 572
 573             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm256_mul_pd(fscal,dx10);
 577             ty               = _mm256_mul_pd(fscal,dy10);
 578             tz               = _mm256_mul_pd(fscal,dz10);
 579
 580             /* Update vectorial force */
 581             fix1             = _mm256_add_pd(fix1,tx);
 582             fiy1             = _mm256_add_pd(fiy1,ty);
 583             fiz1             = _mm256_add_pd(fiz1,tz);
 584
 585             fjx0             = _mm256_add_pd(fjx0,tx);
 586             fjy0             = _mm256_add_pd(fjy0,ty);
 587             fjz0             = _mm256_add_pd(fjz0,tz);
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             r20              = _mm256_mul_pd(rsq20,rinv20);
 594             r20              = _mm256_andnot_pd(dummy_mask,r20);
 595
 596             /* Compute parameters for interactions between i and j atoms */
 597             qq20             = _mm256_mul_pd(iq2,jq0);
 598
 599             /* EWALD ELECTROSTATICS */
 600
 601             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 602             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 603             ewitab           = _mm256_cvttpd_epi32(ewrt);
 604             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 605             ewitab           = _mm_slli_epi32(ewitab,2);
 606             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 607             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 608             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 609             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 610             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 611             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 612             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 613             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 614             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 615
 616             /* Update potential sum for this i atom from the interaction with this j atom. */
 617             velec            = _mm256_andnot_pd(dummy_mask,velec);
 618             velecsum         = _mm256_add_pd(velecsum,velec);
 619
 620             fscal            = felec;
 621
 622             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 623
 624             /* Calculate temporary vectorial force */
 625             tx               = _mm256_mul_pd(fscal,dx20);
 626             ty               = _mm256_mul_pd(fscal,dy20);
 627             tz               = _mm256_mul_pd(fscal,dz20);
 628
 629             /* Update vectorial force */
 630             fix2             = _mm256_add_pd(fix2,tx);
 631             fiy2             = _mm256_add_pd(fiy2,ty);
 632             fiz2             = _mm256_add_pd(fiz2,tz);
 633
 634             fjx0             = _mm256_add_pd(fjx0,tx);
 635             fjy0             = _mm256_add_pd(fjy0,ty);
 636             fjz0             = _mm256_add_pd(fjz0,tz);
 637
 638             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 639             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 640             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 641             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 642
 643             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 644
 645             /* Inner loop uses 141 flops */
 646         }
 647
 648         /* End of innermost loop */
 649
 650         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 651                                                  f+i_coord_offset,fshift+i_shift_offset);
 652
 653         ggid                        = gid[iidx];
 654         /* Update potential energies */
 655         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 656         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 657
 658         /* Increment number of inner iterations */
 659         inneriter                  += j_index_end - j_index_start;
 660
 661         /* Outer loop uses 20 flops */
 662     }
 663
 664     /* Increment number of outer iterations */
 665     outeriter        += nri;
 666
 667     /* Update outer/inner flops */
 668
 669     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*141);
 670 }
 671 /*
 672  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_double
 673  * Electrostatics interaction: Ewald
 674  * VdW interaction:            LennardJones
 675  * Geometry:                   Water3-Particle
 676  * Calculate force/pot:        Force
 677  */
 678 void
 679 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_256_double
 680                     (t_nblist                    * gmx_restrict       nlist,
 681                      rvec                        * gmx_restrict          xx,
 682                      rvec                        * gmx_restrict          ff,
 683                      t_forcerec                  * gmx_restrict          fr,
 684                      t_mdatoms                   * gmx_restrict     mdatoms,
 685                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 686                      t_nrnb                      * gmx_restrict        nrnb)
 687 {
 688     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 689      * just 0 for non-waters.
 690      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 691      * jnr indices corresponding to data put in the four positions in the SIMD register.
 692      */
 693     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 694     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 695     int              jnrA,jnrB,jnrC,jnrD;
 696     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 697     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 698     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 699     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 700     real             rcutoff_scalar;
 701     real             *shiftvec,*fshift,*x,*f;
 702     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 703     real             scratch[4*DIM];
 704     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 705     real *           vdwioffsetptr0;
 706     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 707     real *           vdwioffsetptr1;
 708     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 709     real *           vdwioffsetptr2;
 710     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 711     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 712     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 713     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 714     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 715     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 716     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 717     real             *charge;
 718     int              nvdwtype;
 719     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 720     int              *vdwtype;
 721     real             *vdwparam;
 722     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 723     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 724     __m128i          ewitab;
 725     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 726     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 727     real             *ewtab;
 728     __m256d          dummy_mask,cutoff_mask;
 729     __m128           tmpmask0,tmpmask1;
 730     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 731     __m256d          one     = _mm256_set1_pd(1.0);
 732     __m256d          two     = _mm256_set1_pd(2.0);
 733     x                = xx[0];
 734     f                = ff[0];
 735
 736     nri              = nlist->nri;
 737     iinr             = nlist->iinr;
 738     jindex           = nlist->jindex;
 739     jjnr             = nlist->jjnr;
 740     shiftidx         = nlist->shift;
 741     gid              = nlist->gid;
 742     shiftvec         = fr->shift_vec[0];
 743     fshift           = fr->fshift[0];
 744     facel            = _mm256_set1_pd(fr->epsfac);
 745     charge           = mdatoms->chargeA;
 746     nvdwtype         = fr->ntype;
 747     vdwparam         = fr->nbfp;
 748     vdwtype          = mdatoms->typeA;
 749
 750     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 751     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 752     beta2            = _mm256_mul_pd(beta,beta);
 753     beta3            = _mm256_mul_pd(beta,beta2);
 754
 755     ewtab            = fr->ic->tabq_coul_F;
 756     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 757     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 758
 759     /* Setup water-specific parameters */
 760     inr              = nlist->iinr[0];
 761     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 762     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 763     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 764     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 765
 766     /* Avoid stupid compiler warnings */
 767     jnrA = jnrB = jnrC = jnrD = 0;
 768     j_coord_offsetA = 0;
 769     j_coord_offsetB = 0;
 770     j_coord_offsetC = 0;
 771     j_coord_offsetD = 0;
 772
 773     outeriter        = 0;
 774     inneriter        = 0;
 775
 776     for(iidx=0;iidx<4*DIM;iidx++)
 777     {
 778         scratch[iidx] = 0.0;
 779     }
 780
 781     /* Start outer loop over neighborlists */
 782     for(iidx=0; iidx<nri; iidx++)
 783     {
 784         /* Load shift vector for this list */
 785         i_shift_offset   = DIM*shiftidx[iidx];
 786
 787         /* Load limits for loop over neighbors */
 788         j_index_start    = jindex[iidx];
 789         j_index_end      = jindex[iidx+1];
 790
 791         /* Get outer coordinate index */
 792         inr              = iinr[iidx];
 793         i_coord_offset   = DIM*inr;
 794
 795         /* Load i particle coords and add shift vector */
 796         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 797                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 798
 799         fix0             = _mm256_setzero_pd();
 800         fiy0             = _mm256_setzero_pd();
 801         fiz0             = _mm256_setzero_pd();
 802         fix1             = _mm256_setzero_pd();
 803         fiy1             = _mm256_setzero_pd();
 804         fiz1             = _mm256_setzero_pd();
 805         fix2             = _mm256_setzero_pd();
 806         fiy2             = _mm256_setzero_pd();
 807         fiz2             = _mm256_setzero_pd();
 808
 809         /* Start inner kernel loop */
 810         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 811         {
 812
 813             /* Get j neighbor index, and coordinate index */
 814             jnrA             = jjnr[jidx];
 815             jnrB             = jjnr[jidx+1];
 816             jnrC             = jjnr[jidx+2];
 817             jnrD             = jjnr[jidx+3];
 818             j_coord_offsetA  = DIM*jnrA;
 819             j_coord_offsetB  = DIM*jnrB;
 820             j_coord_offsetC  = DIM*jnrC;
 821             j_coord_offsetD  = DIM*jnrD;
 822
 823             /* load j atom coordinates */
 824             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 825                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 826                                                  &jx0,&jy0,&jz0);
 827
 828             /* Calculate displacement vector */
 829             dx00             = _mm256_sub_pd(ix0,jx0);
 830             dy00             = _mm256_sub_pd(iy0,jy0);
 831             dz00             = _mm256_sub_pd(iz0,jz0);
 832             dx10             = _mm256_sub_pd(ix1,jx0);
 833             dy10             = _mm256_sub_pd(iy1,jy0);
 834             dz10             = _mm256_sub_pd(iz1,jz0);
 835             dx20             = _mm256_sub_pd(ix2,jx0);
 836             dy20             = _mm256_sub_pd(iy2,jy0);
 837             dz20             = _mm256_sub_pd(iz2,jz0);
 838
 839             /* Calculate squared distance and things based on it */
 840             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 841             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 842             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 843
 844             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 845             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 846             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 847
 848             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 849             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 850             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 851
 852             /* Load parameters for j particles */
 853             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 854                                                                  charge+jnrC+0,charge+jnrD+0);
 855             vdwjidx0A        = 2*vdwtype[jnrA+0];
 856             vdwjidx0B        = 2*vdwtype[jnrB+0];
 857             vdwjidx0C        = 2*vdwtype[jnrC+0];
 858             vdwjidx0D        = 2*vdwtype[jnrD+0];
 859
 860             fjx0             = _mm256_setzero_pd();
 861             fjy0             = _mm256_setzero_pd();
 862             fjz0             = _mm256_setzero_pd();
 863
 864             /**************************
 865              * CALCULATE INTERACTIONS *
 866              **************************/
 867
 868             r00              = _mm256_mul_pd(rsq00,rinv00);
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq00             = _mm256_mul_pd(iq0,jq0);
 872             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 873                                             vdwioffsetptr0+vdwjidx0B,
 874                                             vdwioffsetptr0+vdwjidx0C,
 875                                             vdwioffsetptr0+vdwjidx0D,
 876                                             &c6_00,&c12_00);
 877
 878             /* EWALD ELECTROSTATICS */
 879
 880             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 881             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 882             ewitab           = _mm256_cvttpd_epi32(ewrt);
 883             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 884             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 885                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 886                                             &ewtabF,&ewtabFn);
 887             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 888             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 889
 890             /* LENNARD-JONES DISPERSION/REPULSION */
 891
 892             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 893             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 894
 895             fscal            = _mm256_add_pd(felec,fvdw);
 896
 897             /* Calculate temporary vectorial force */
 898             tx               = _mm256_mul_pd(fscal,dx00);
 899             ty               = _mm256_mul_pd(fscal,dy00);
 900             tz               = _mm256_mul_pd(fscal,dz00);
 901
 902             /* Update vectorial force */
 903             fix0             = _mm256_add_pd(fix0,tx);
 904             fiy0             = _mm256_add_pd(fiy0,ty);
 905             fiz0             = _mm256_add_pd(fiz0,tz);
 906
 907             fjx0             = _mm256_add_pd(fjx0,tx);
 908             fjy0             = _mm256_add_pd(fjy0,ty);
 909             fjz0             = _mm256_add_pd(fjz0,tz);
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             r10              = _mm256_mul_pd(rsq10,rinv10);
 916
 917             /* Compute parameters for interactions between i and j atoms */
 918             qq10             = _mm256_mul_pd(iq1,jq0);
 919
 920             /* EWALD ELECTROSTATICS */
 921
 922             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 923             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 924             ewitab           = _mm256_cvttpd_epi32(ewrt);
 925             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 926             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 927                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 928                                             &ewtabF,&ewtabFn);
 929             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 930             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 931
 932             fscal            = felec;
 933
 934             /* Calculate temporary vectorial force */
 935             tx               = _mm256_mul_pd(fscal,dx10);
 936             ty               = _mm256_mul_pd(fscal,dy10);
 937             tz               = _mm256_mul_pd(fscal,dz10);
 938
 939             /* Update vectorial force */
 940             fix1             = _mm256_add_pd(fix1,tx);
 941             fiy1             = _mm256_add_pd(fiy1,ty);
 942             fiz1             = _mm256_add_pd(fiz1,tz);
 943
 944             fjx0             = _mm256_add_pd(fjx0,tx);
 945             fjy0             = _mm256_add_pd(fjy0,ty);
 946             fjz0             = _mm256_add_pd(fjz0,tz);
 947
 948             /**************************
 949              * CALCULATE INTERACTIONS *
 950              **************************/
 951
 952             r20              = _mm256_mul_pd(rsq20,rinv20);
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             qq20             = _mm256_mul_pd(iq2,jq0);
 956
 957             /* EWALD ELECTROSTATICS */
 958
 959             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 960             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 961             ewitab           = _mm256_cvttpd_epi32(ewrt);
 962             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 963             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 964                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 965                                             &ewtabF,&ewtabFn);
 966             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 967             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 968
 969             fscal            = felec;
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm256_mul_pd(fscal,dx20);
 973             ty               = _mm256_mul_pd(fscal,dy20);
 974             tz               = _mm256_mul_pd(fscal,dz20);
 975
 976             /* Update vectorial force */
 977             fix2             = _mm256_add_pd(fix2,tx);
 978             fiy2             = _mm256_add_pd(fiy2,ty);
 979             fiz2             = _mm256_add_pd(fiz2,tz);
 980
 981             fjx0             = _mm256_add_pd(fjx0,tx);
 982             fjy0             = _mm256_add_pd(fjy0,ty);
 983             fjz0             = _mm256_add_pd(fjz0,tz);
 984
 985             fjptrA             = f+j_coord_offsetA;
 986             fjptrB             = f+j_coord_offsetB;
 987             fjptrC             = f+j_coord_offsetC;
 988             fjptrD             = f+j_coord_offsetD;
 989
 990             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 991
 992             /* Inner loop uses 118 flops */
 993         }
 994
 995         if(jidx<j_index_end)
 996         {
 997
 998             /* Get j neighbor index, and coordinate index */
 999             jnrlistA         = jjnr[jidx];
1000             jnrlistB         = jjnr[jidx+1];
1001             jnrlistC         = jjnr[jidx+2];
1002             jnrlistD         = jjnr[jidx+3];
1003             /* Sign of each element will be negative for non-real atoms.
1004              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1005              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1006              */
1007             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1008
1009             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1010             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1011             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1012
1013             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1014             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1015             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1016             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1017             j_coord_offsetA  = DIM*jnrA;
1018             j_coord_offsetB  = DIM*jnrB;
1019             j_coord_offsetC  = DIM*jnrC;
1020             j_coord_offsetD  = DIM*jnrD;
1021
1022             /* load j atom coordinates */
1023             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1024                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1025                                                  &jx0,&jy0,&jz0);
1026
1027             /* Calculate displacement vector */
1028             dx00             = _mm256_sub_pd(ix0,jx0);
1029             dy00             = _mm256_sub_pd(iy0,jy0);
1030             dz00             = _mm256_sub_pd(iz0,jz0);
1031             dx10             = _mm256_sub_pd(ix1,jx0);
1032             dy10             = _mm256_sub_pd(iy1,jy0);
1033             dz10             = _mm256_sub_pd(iz1,jz0);
1034             dx20             = _mm256_sub_pd(ix2,jx0);
1035             dy20             = _mm256_sub_pd(iy2,jy0);
1036             dz20             = _mm256_sub_pd(iz2,jz0);
1037
1038             /* Calculate squared distance and things based on it */
1039             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1040             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1041             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1042
1043             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1044             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1045             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1046
1047             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1048             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1049             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1050
1051             /* Load parameters for j particles */
1052             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1053                                                                  charge+jnrC+0,charge+jnrD+0);
1054             vdwjidx0A        = 2*vdwtype[jnrA+0];
1055             vdwjidx0B        = 2*vdwtype[jnrB+0];
1056             vdwjidx0C        = 2*vdwtype[jnrC+0];
1057             vdwjidx0D        = 2*vdwtype[jnrD+0];
1058
1059             fjx0             = _mm256_setzero_pd();
1060             fjy0             = _mm256_setzero_pd();
1061             fjz0             = _mm256_setzero_pd();
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             r00              = _mm256_mul_pd(rsq00,rinv00);
1068             r00              = _mm256_andnot_pd(dummy_mask,r00);
1069
1070             /* Compute parameters for interactions between i and j atoms */
1071             qq00             = _mm256_mul_pd(iq0,jq0);
1072             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1073                                             vdwioffsetptr0+vdwjidx0B,
1074                                             vdwioffsetptr0+vdwjidx0C,
1075                                             vdwioffsetptr0+vdwjidx0D,
1076                                             &c6_00,&c12_00);
1077
1078             /* EWALD ELECTROSTATICS */
1079
1080             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1081             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1082             ewitab           = _mm256_cvttpd_epi32(ewrt);
1083             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1084             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1085                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1086                                             &ewtabF,&ewtabFn);
1087             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1088             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1089
1090             /* LENNARD-JONES DISPERSION/REPULSION */
1091
1092             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1093             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1094
1095             fscal            = _mm256_add_pd(felec,fvdw);
1096
1097             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1098
1099             /* Calculate temporary vectorial force */
1100             tx               = _mm256_mul_pd(fscal,dx00);
1101             ty               = _mm256_mul_pd(fscal,dy00);
1102             tz               = _mm256_mul_pd(fscal,dz00);
1103
1104             /* Update vectorial force */
1105             fix0             = _mm256_add_pd(fix0,tx);
1106             fiy0             = _mm256_add_pd(fiy0,ty);
1107             fiz0             = _mm256_add_pd(fiz0,tz);
1108
1109             fjx0             = _mm256_add_pd(fjx0,tx);
1110             fjy0             = _mm256_add_pd(fjy0,ty);
1111             fjz0             = _mm256_add_pd(fjz0,tz);
1112
1113             /**************************
1114              * CALCULATE INTERACTIONS *
1115              **************************/
1116
1117             r10              = _mm256_mul_pd(rsq10,rinv10);
1118             r10              = _mm256_andnot_pd(dummy_mask,r10);
1119
1120             /* Compute parameters for interactions between i and j atoms */
1121             qq10             = _mm256_mul_pd(iq1,jq0);
1122
1123             /* EWALD ELECTROSTATICS */
1124
1125             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1126             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1127             ewitab           = _mm256_cvttpd_epi32(ewrt);
1128             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1129             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1130                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1131                                             &ewtabF,&ewtabFn);
1132             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1133             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1134
1135             fscal            = felec;
1136
1137             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm256_mul_pd(fscal,dx10);
1141             ty               = _mm256_mul_pd(fscal,dy10);
1142             tz               = _mm256_mul_pd(fscal,dz10);
1143
1144             /* Update vectorial force */
1145             fix1             = _mm256_add_pd(fix1,tx);
1146             fiy1             = _mm256_add_pd(fiy1,ty);
1147             fiz1             = _mm256_add_pd(fiz1,tz);
1148
1149             fjx0             = _mm256_add_pd(fjx0,tx);
1150             fjy0             = _mm256_add_pd(fjy0,ty);
1151             fjz0             = _mm256_add_pd(fjz0,tz);
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             r20              = _mm256_mul_pd(rsq20,rinv20);
1158             r20              = _mm256_andnot_pd(dummy_mask,r20);
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             qq20             = _mm256_mul_pd(iq2,jq0);
1162
1163             /* EWALD ELECTROSTATICS */
1164
1165             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1166             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1167             ewitab           = _mm256_cvttpd_epi32(ewrt);
1168             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1169             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1170                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1171                                             &ewtabF,&ewtabFn);
1172             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1173             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1174
1175             fscal            = felec;
1176
1177             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1178
1179             /* Calculate temporary vectorial force */
1180             tx               = _mm256_mul_pd(fscal,dx20);
1181             ty               = _mm256_mul_pd(fscal,dy20);
1182             tz               = _mm256_mul_pd(fscal,dz20);
1183
1184             /* Update vectorial force */
1185             fix2             = _mm256_add_pd(fix2,tx);
1186             fiy2             = _mm256_add_pd(fiy2,ty);
1187             fiz2             = _mm256_add_pd(fiz2,tz);
1188
1189             fjx0             = _mm256_add_pd(fjx0,tx);
1190             fjy0             = _mm256_add_pd(fjy0,ty);
1191             fjz0             = _mm256_add_pd(fjz0,tz);
1192
1193             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1194             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1195             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1196             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1197
1198             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1199
1200             /* Inner loop uses 121 flops */
1201         }
1202
1203         /* End of innermost loop */
1204
1205         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1206                                                  f+i_coord_offset,fshift+i_shift_offset);
1207
1208         /* Increment number of inner iterations */
1209         inneriter                  += j_index_end - j_index_start;
1210
1211         /* Outer loop uses 18 flops */
1212     }
1213
1214     /* Increment number of outer iterations */
1215     outeriter        += nri;
1216
1217     /* Update outer/inner flops */
1218
1219     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*121);
1220 }