src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwNone_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_ElecCSTab_VdwNone_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwNone_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     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m256d          dummy_mask,cutoff_mask;
 104     __m128           tmpmask0,tmpmask1;
 105     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 106     __m256d          one     = _mm256_set1_pd(1.0);
 107     __m256d          two     = _mm256_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121
 122     vftab            = kernel_data->table_elec->data;
 123     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 128     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 129     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 163
 164         fix0             = _mm256_setzero_pd();
 165         fiy0             = _mm256_setzero_pd();
 166         fiz0             = _mm256_setzero_pd();
 167         fix1             = _mm256_setzero_pd();
 168         fiy1             = _mm256_setzero_pd();
 169         fiz1             = _mm256_setzero_pd();
 170         fix2             = _mm256_setzero_pd();
 171         fiy2             = _mm256_setzero_pd();
 172         fiz2             = _mm256_setzero_pd();
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm256_setzero_pd();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             jnrC             = jjnr[jidx+2];
 185             jnrD             = jjnr[jidx+3];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190
 191             /* load j atom coordinates */
 192             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 194                                                  &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm256_sub_pd(ix0,jx0);
 198             dy00             = _mm256_sub_pd(iy0,jy0);
 199             dz00             = _mm256_sub_pd(iz0,jz0);
 200             dx10             = _mm256_sub_pd(ix1,jx0);
 201             dy10             = _mm256_sub_pd(iy1,jy0);
 202             dz10             = _mm256_sub_pd(iz1,jz0);
 203             dx20             = _mm256_sub_pd(ix2,jx0);
 204             dy20             = _mm256_sub_pd(iy2,jy0);
 205             dz20             = _mm256_sub_pd(iz2,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 209             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 210             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 211
 212             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 213             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 214             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 215
 216             /* Load parameters for j particles */
 217             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 218                                                                  charge+jnrC+0,charge+jnrD+0);
 219
 220             fjx0             = _mm256_setzero_pd();
 221             fjy0             = _mm256_setzero_pd();
 222             fjz0             = _mm256_setzero_pd();
 223
 224             /**************************
 225              * CALCULATE INTERACTIONS *
 226              **************************/
 227
 228             r00              = _mm256_mul_pd(rsq00,rinv00);
 229
 230             /* Compute parameters for interactions between i and j atoms */
 231             qq00             = _mm256_mul_pd(iq0,jq0);
 232
 233             /* Calculate table index by multiplying r with table scale and truncate to integer */
 234             rt               = _mm256_mul_pd(r00,vftabscale);
 235             vfitab           = _mm256_cvttpd_epi32(rt);
 236             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 237             vfitab           = _mm_slli_epi32(vfitab,2);
 238
 239             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 240             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 241             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 242             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 243             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 244             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 245             Heps             = _mm256_mul_pd(vfeps,H);
 246             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 247             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 248             velec            = _mm256_mul_pd(qq00,VV);
 249             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 250             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             velecsum         = _mm256_add_pd(velecsum,velec);
 254
 255             fscal            = felec;
 256
 257             /* Calculate temporary vectorial force */
 258             tx               = _mm256_mul_pd(fscal,dx00);
 259             ty               = _mm256_mul_pd(fscal,dy00);
 260             tz               = _mm256_mul_pd(fscal,dz00);
 261
 262             /* Update vectorial force */
 263             fix0             = _mm256_add_pd(fix0,tx);
 264             fiy0             = _mm256_add_pd(fiy0,ty);
 265             fiz0             = _mm256_add_pd(fiz0,tz);
 266
 267             fjx0             = _mm256_add_pd(fjx0,tx);
 268             fjy0             = _mm256_add_pd(fjy0,ty);
 269             fjz0             = _mm256_add_pd(fjz0,tz);
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             r10              = _mm256_mul_pd(rsq10,rinv10);
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             qq10             = _mm256_mul_pd(iq1,jq0);
 279
 280             /* Calculate table index by multiplying r with table scale and truncate to integer */
 281             rt               = _mm256_mul_pd(r10,vftabscale);
 282             vfitab           = _mm256_cvttpd_epi32(rt);
 283             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 284             vfitab           = _mm_slli_epi32(vfitab,2);
 285
 286             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 287             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 288             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 289             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 290             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 291             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 292             Heps             = _mm256_mul_pd(vfeps,H);
 293             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 294             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 295             velec            = _mm256_mul_pd(qq10,VV);
 296             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 297             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velecsum         = _mm256_add_pd(velecsum,velec);
 301
 302             fscal            = felec;
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm256_mul_pd(fscal,dx10);
 306             ty               = _mm256_mul_pd(fscal,dy10);
 307             tz               = _mm256_mul_pd(fscal,dz10);
 308
 309             /* Update vectorial force */
 310             fix1             = _mm256_add_pd(fix1,tx);
 311             fiy1             = _mm256_add_pd(fiy1,ty);
 312             fiz1             = _mm256_add_pd(fiz1,tz);
 313
 314             fjx0             = _mm256_add_pd(fjx0,tx);
 315             fjy0             = _mm256_add_pd(fjy0,ty);
 316             fjz0             = _mm256_add_pd(fjz0,tz);
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             r20              = _mm256_mul_pd(rsq20,rinv20);
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq20             = _mm256_mul_pd(iq2,jq0);
 326
 327             /* Calculate table index by multiplying r with table scale and truncate to integer */
 328             rt               = _mm256_mul_pd(r20,vftabscale);
 329             vfitab           = _mm256_cvttpd_epi32(rt);
 330             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 331             vfitab           = _mm_slli_epi32(vfitab,2);
 332
 333             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 334             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 335             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 336             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 337             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 338             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 339             Heps             = _mm256_mul_pd(vfeps,H);
 340             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 341             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 342             velec            = _mm256_mul_pd(qq20,VV);
 343             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 344             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm256_add_pd(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm256_mul_pd(fscal,dx20);
 353             ty               = _mm256_mul_pd(fscal,dy20);
 354             tz               = _mm256_mul_pd(fscal,dz20);
 355
 356             /* Update vectorial force */
 357             fix2             = _mm256_add_pd(fix2,tx);
 358             fiy2             = _mm256_add_pd(fiy2,ty);
 359             fiz2             = _mm256_add_pd(fiz2,tz);
 360
 361             fjx0             = _mm256_add_pd(fjx0,tx);
 362             fjy0             = _mm256_add_pd(fjy0,ty);
 363             fjz0             = _mm256_add_pd(fjz0,tz);
 364
 365             fjptrA             = f+j_coord_offsetA;
 366             fjptrB             = f+j_coord_offsetB;
 367             fjptrC             = f+j_coord_offsetC;
 368             fjptrD             = f+j_coord_offsetD;
 369
 370             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 371
 372             /* Inner loop uses 132 flops */
 373         }
 374
 375         if(jidx<j_index_end)
 376         {
 377
 378             /* Get j neighbor index, and coordinate index */
 379             jnrlistA         = jjnr[jidx];
 380             jnrlistB         = jjnr[jidx+1];
 381             jnrlistC         = jjnr[jidx+2];
 382             jnrlistD         = jjnr[jidx+3];
 383             /* Sign of each element will be negative for non-real atoms.
 384              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 385              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 386              */
 387             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 388
 389             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 390             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 391             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 392
 393             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 394             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 395             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 396             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 397             j_coord_offsetA  = DIM*jnrA;
 398             j_coord_offsetB  = DIM*jnrB;
 399             j_coord_offsetC  = DIM*jnrC;
 400             j_coord_offsetD  = DIM*jnrD;
 401
 402             /* load j atom coordinates */
 403             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 404                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 405                                                  &jx0,&jy0,&jz0);
 406
 407             /* Calculate displacement vector */
 408             dx00             = _mm256_sub_pd(ix0,jx0);
 409             dy00             = _mm256_sub_pd(iy0,jy0);
 410             dz00             = _mm256_sub_pd(iz0,jz0);
 411             dx10             = _mm256_sub_pd(ix1,jx0);
 412             dy10             = _mm256_sub_pd(iy1,jy0);
 413             dz10             = _mm256_sub_pd(iz1,jz0);
 414             dx20             = _mm256_sub_pd(ix2,jx0);
 415             dy20             = _mm256_sub_pd(iy2,jy0);
 416             dz20             = _mm256_sub_pd(iz2,jz0);
 417
 418             /* Calculate squared distance and things based on it */
 419             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 420             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 421             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 422
 423             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 424             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 425             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 426
 427             /* Load parameters for j particles */
 428             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 429                                                                  charge+jnrC+0,charge+jnrD+0);
 430
 431             fjx0             = _mm256_setzero_pd();
 432             fjy0             = _mm256_setzero_pd();
 433             fjz0             = _mm256_setzero_pd();
 434
 435             /**************************
 436              * CALCULATE INTERACTIONS *
 437              **************************/
 438
 439             r00              = _mm256_mul_pd(rsq00,rinv00);
 440             r00              = _mm256_andnot_pd(dummy_mask,r00);
 441
 442             /* Compute parameters for interactions between i and j atoms */
 443             qq00             = _mm256_mul_pd(iq0,jq0);
 444
 445             /* Calculate table index by multiplying r with table scale and truncate to integer */
 446             rt               = _mm256_mul_pd(r00,vftabscale);
 447             vfitab           = _mm256_cvttpd_epi32(rt);
 448             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 449             vfitab           = _mm_slli_epi32(vfitab,2);
 450
 451             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 452             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 453             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 454             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 455             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 456             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 457             Heps             = _mm256_mul_pd(vfeps,H);
 458             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 459             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 460             velec            = _mm256_mul_pd(qq00,VV);
 461             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 462             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 463
 464             /* Update potential sum for this i atom from the interaction with this j atom. */
 465             velec            = _mm256_andnot_pd(dummy_mask,velec);
 466             velecsum         = _mm256_add_pd(velecsum,velec);
 467
 468             fscal            = felec;
 469
 470             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 471
 472             /* Calculate temporary vectorial force */
 473             tx               = _mm256_mul_pd(fscal,dx00);
 474             ty               = _mm256_mul_pd(fscal,dy00);
 475             tz               = _mm256_mul_pd(fscal,dz00);
 476
 477             /* Update vectorial force */
 478             fix0             = _mm256_add_pd(fix0,tx);
 479             fiy0             = _mm256_add_pd(fiy0,ty);
 480             fiz0             = _mm256_add_pd(fiz0,tz);
 481
 482             fjx0             = _mm256_add_pd(fjx0,tx);
 483             fjy0             = _mm256_add_pd(fjy0,ty);
 484             fjz0             = _mm256_add_pd(fjz0,tz);
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             r10              = _mm256_mul_pd(rsq10,rinv10);
 491             r10              = _mm256_andnot_pd(dummy_mask,r10);
 492
 493             /* Compute parameters for interactions between i and j atoms */
 494             qq10             = _mm256_mul_pd(iq1,jq0);
 495
 496             /* Calculate table index by multiplying r with table scale and truncate to integer */
 497             rt               = _mm256_mul_pd(r10,vftabscale);
 498             vfitab           = _mm256_cvttpd_epi32(rt);
 499             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 500             vfitab           = _mm_slli_epi32(vfitab,2);
 501
 502             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 503             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 504             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 505             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 506             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 507             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 508             Heps             = _mm256_mul_pd(vfeps,H);
 509             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 510             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 511             velec            = _mm256_mul_pd(qq10,VV);
 512             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 513             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 514
 515             /* Update potential sum for this i atom from the interaction with this j atom. */
 516             velec            = _mm256_andnot_pd(dummy_mask,velec);
 517             velecsum         = _mm256_add_pd(velecsum,velec);
 518
 519             fscal            = felec;
 520
 521             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 522
 523             /* Calculate temporary vectorial force */
 524             tx               = _mm256_mul_pd(fscal,dx10);
 525             ty               = _mm256_mul_pd(fscal,dy10);
 526             tz               = _mm256_mul_pd(fscal,dz10);
 527
 528             /* Update vectorial force */
 529             fix1             = _mm256_add_pd(fix1,tx);
 530             fiy1             = _mm256_add_pd(fiy1,ty);
 531             fiz1             = _mm256_add_pd(fiz1,tz);
 532
 533             fjx0             = _mm256_add_pd(fjx0,tx);
 534             fjy0             = _mm256_add_pd(fjy0,ty);
 535             fjz0             = _mm256_add_pd(fjz0,tz);
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             r20              = _mm256_mul_pd(rsq20,rinv20);
 542             r20              = _mm256_andnot_pd(dummy_mask,r20);
 543
 544             /* Compute parameters for interactions between i and j atoms */
 545             qq20             = _mm256_mul_pd(iq2,jq0);
 546
 547             /* Calculate table index by multiplying r with table scale and truncate to integer */
 548             rt               = _mm256_mul_pd(r20,vftabscale);
 549             vfitab           = _mm256_cvttpd_epi32(rt);
 550             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 551             vfitab           = _mm_slli_epi32(vfitab,2);
 552
 553             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 554             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 555             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 556             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 557             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 558             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 559             Heps             = _mm256_mul_pd(vfeps,H);
 560             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 561             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 562             velec            = _mm256_mul_pd(qq20,VV);
 563             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 564             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 565
 566             /* Update potential sum for this i atom from the interaction with this j atom. */
 567             velec            = _mm256_andnot_pd(dummy_mask,velec);
 568             velecsum         = _mm256_add_pd(velecsum,velec);
 569
 570             fscal            = felec;
 571
 572             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 573
 574             /* Calculate temporary vectorial force */
 575             tx               = _mm256_mul_pd(fscal,dx20);
 576             ty               = _mm256_mul_pd(fscal,dy20);
 577             tz               = _mm256_mul_pd(fscal,dz20);
 578
 579             /* Update vectorial force */
 580             fix2             = _mm256_add_pd(fix2,tx);
 581             fiy2             = _mm256_add_pd(fiy2,ty);
 582             fiz2             = _mm256_add_pd(fiz2,tz);
 583
 584             fjx0             = _mm256_add_pd(fjx0,tx);
 585             fjy0             = _mm256_add_pd(fjy0,ty);
 586             fjz0             = _mm256_add_pd(fjz0,tz);
 587
 588             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 589             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 590             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 591             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 592
 593             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 594
 595             /* Inner loop uses 135 flops */
 596         }
 597
 598         /* End of innermost loop */
 599
 600         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 601                                                  f+i_coord_offset,fshift+i_shift_offset);
 602
 603         ggid                        = gid[iidx];
 604         /* Update potential energies */
 605         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 606
 607         /* Increment number of inner iterations */
 608         inneriter                  += j_index_end - j_index_start;
 609
 610         /* Outer loop uses 19 flops */
 611     }
 612
 613     /* Increment number of outer iterations */
 614     outeriter        += nri;
 615
 616     /* Update outer/inner flops */
 617
 618     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*135);
 619 }
 620 /*
 621  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_avx_256_double
 622  * Electrostatics interaction: CubicSplineTable
 623  * VdW interaction:            None
 624  * Geometry:                   Water3-Particle
 625  * Calculate force/pot:        Force
 626  */
 627 void
 628 nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_avx_256_double
 629                     (t_nblist                    * gmx_restrict       nlist,
 630                      rvec                        * gmx_restrict          xx,
 631                      rvec                        * gmx_restrict          ff,
 632                      t_forcerec                  * gmx_restrict          fr,
 633                      t_mdatoms                   * gmx_restrict     mdatoms,
 634                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 635                      t_nrnb                      * gmx_restrict        nrnb)
 636 {
 637     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 638      * just 0 for non-waters.
 639      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 640      * jnr indices corresponding to data put in the four positions in the SIMD register.
 641      */
 642     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 643     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 644     int              jnrA,jnrB,jnrC,jnrD;
 645     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 646     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 647     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 648     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 649     real             rcutoff_scalar;
 650     real             *shiftvec,*fshift,*x,*f;
 651     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 652     real             scratch[4*DIM];
 653     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 654     real *           vdwioffsetptr0;
 655     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 656     real *           vdwioffsetptr1;
 657     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 658     real *           vdwioffsetptr2;
 659     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 660     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 661     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 662     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 663     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 664     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 665     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 666     real             *charge;
 667     __m128i          vfitab;
 668     __m128i          ifour       = _mm_set1_epi32(4);
 669     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 670     real             *vftab;
 671     __m256d          dummy_mask,cutoff_mask;
 672     __m128           tmpmask0,tmpmask1;
 673     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 674     __m256d          one     = _mm256_set1_pd(1.0);
 675     __m256d          two     = _mm256_set1_pd(2.0);
 676     x                = xx[0];
 677     f                = ff[0];
 678
 679     nri              = nlist->nri;
 680     iinr             = nlist->iinr;
 681     jindex           = nlist->jindex;
 682     jjnr             = nlist->jjnr;
 683     shiftidx         = nlist->shift;
 684     gid              = nlist->gid;
 685     shiftvec         = fr->shift_vec[0];
 686     fshift           = fr->fshift[0];
 687     facel            = _mm256_set1_pd(fr->epsfac);
 688     charge           = mdatoms->chargeA;
 689
 690     vftab            = kernel_data->table_elec->data;
 691     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 692
 693     /* Setup water-specific parameters */
 694     inr              = nlist->iinr[0];
 695     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 696     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 697     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 698
 699     /* Avoid stupid compiler warnings */
 700     jnrA = jnrB = jnrC = jnrD = 0;
 701     j_coord_offsetA = 0;
 702     j_coord_offsetB = 0;
 703     j_coord_offsetC = 0;
 704     j_coord_offsetD = 0;
 705
 706     outeriter        = 0;
 707     inneriter        = 0;
 708
 709     for(iidx=0;iidx<4*DIM;iidx++)
 710     {
 711         scratch[iidx] = 0.0;
 712     }
 713
 714     /* Start outer loop over neighborlists */
 715     for(iidx=0; iidx<nri; iidx++)
 716     {
 717         /* Load shift vector for this list */
 718         i_shift_offset   = DIM*shiftidx[iidx];
 719
 720         /* Load limits for loop over neighbors */
 721         j_index_start    = jindex[iidx];
 722         j_index_end      = jindex[iidx+1];
 723
 724         /* Get outer coordinate index */
 725         inr              = iinr[iidx];
 726         i_coord_offset   = DIM*inr;
 727
 728         /* Load i particle coords and add shift vector */
 729         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 730                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 731
 732         fix0             = _mm256_setzero_pd();
 733         fiy0             = _mm256_setzero_pd();
 734         fiz0             = _mm256_setzero_pd();
 735         fix1             = _mm256_setzero_pd();
 736         fiy1             = _mm256_setzero_pd();
 737         fiz1             = _mm256_setzero_pd();
 738         fix2             = _mm256_setzero_pd();
 739         fiy2             = _mm256_setzero_pd();
 740         fiz2             = _mm256_setzero_pd();
 741
 742         /* Start inner kernel loop */
 743         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 744         {
 745
 746             /* Get j neighbor index, and coordinate index */
 747             jnrA             = jjnr[jidx];
 748             jnrB             = jjnr[jidx+1];
 749             jnrC             = jjnr[jidx+2];
 750             jnrD             = jjnr[jidx+3];
 751             j_coord_offsetA  = DIM*jnrA;
 752             j_coord_offsetB  = DIM*jnrB;
 753             j_coord_offsetC  = DIM*jnrC;
 754             j_coord_offsetD  = DIM*jnrD;
 755
 756             /* load j atom coordinates */
 757             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 758                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 759                                                  &jx0,&jy0,&jz0);
 760
 761             /* Calculate displacement vector */
 762             dx00             = _mm256_sub_pd(ix0,jx0);
 763             dy00             = _mm256_sub_pd(iy0,jy0);
 764             dz00             = _mm256_sub_pd(iz0,jz0);
 765             dx10             = _mm256_sub_pd(ix1,jx0);
 766             dy10             = _mm256_sub_pd(iy1,jy0);
 767             dz10             = _mm256_sub_pd(iz1,jz0);
 768             dx20             = _mm256_sub_pd(ix2,jx0);
 769             dy20             = _mm256_sub_pd(iy2,jy0);
 770             dz20             = _mm256_sub_pd(iz2,jz0);
 771
 772             /* Calculate squared distance and things based on it */
 773             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 774             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 775             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 776
 777             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 778             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 779             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 780
 781             /* Load parameters for j particles */
 782             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 783                                                                  charge+jnrC+0,charge+jnrD+0);
 784
 785             fjx0             = _mm256_setzero_pd();
 786             fjy0             = _mm256_setzero_pd();
 787             fjz0             = _mm256_setzero_pd();
 788
 789             /**************************
 790              * CALCULATE INTERACTIONS *
 791              **************************/
 792
 793             r00              = _mm256_mul_pd(rsq00,rinv00);
 794
 795             /* Compute parameters for interactions between i and j atoms */
 796             qq00             = _mm256_mul_pd(iq0,jq0);
 797
 798             /* Calculate table index by multiplying r with table scale and truncate to integer */
 799             rt               = _mm256_mul_pd(r00,vftabscale);
 800             vfitab           = _mm256_cvttpd_epi32(rt);
 801             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 802             vfitab           = _mm_slli_epi32(vfitab,2);
 803
 804             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 805             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 806             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 807             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 808             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 809             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 810             Heps             = _mm256_mul_pd(vfeps,H);
 811             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 812             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 813             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 814
 815             fscal            = felec;
 816
 817             /* Calculate temporary vectorial force */
 818             tx               = _mm256_mul_pd(fscal,dx00);
 819             ty               = _mm256_mul_pd(fscal,dy00);
 820             tz               = _mm256_mul_pd(fscal,dz00);
 821
 822             /* Update vectorial force */
 823             fix0             = _mm256_add_pd(fix0,tx);
 824             fiy0             = _mm256_add_pd(fiy0,ty);
 825             fiz0             = _mm256_add_pd(fiz0,tz);
 826
 827             fjx0             = _mm256_add_pd(fjx0,tx);
 828             fjy0             = _mm256_add_pd(fjy0,ty);
 829             fjz0             = _mm256_add_pd(fjz0,tz);
 830
 831             /**************************
 832              * CALCULATE INTERACTIONS *
 833              **************************/
 834
 835             r10              = _mm256_mul_pd(rsq10,rinv10);
 836
 837             /* Compute parameters for interactions between i and j atoms */
 838             qq10             = _mm256_mul_pd(iq1,jq0);
 839
 840             /* Calculate table index by multiplying r with table scale and truncate to integer */
 841             rt               = _mm256_mul_pd(r10,vftabscale);
 842             vfitab           = _mm256_cvttpd_epi32(rt);
 843             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 844             vfitab           = _mm_slli_epi32(vfitab,2);
 845
 846             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 847             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 848             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 849             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 850             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 851             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 852             Heps             = _mm256_mul_pd(vfeps,H);
 853             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 854             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 855             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 856
 857             fscal            = felec;
 858
 859             /* Calculate temporary vectorial force */
 860             tx               = _mm256_mul_pd(fscal,dx10);
 861             ty               = _mm256_mul_pd(fscal,dy10);
 862             tz               = _mm256_mul_pd(fscal,dz10);
 863
 864             /* Update vectorial force */
 865             fix1             = _mm256_add_pd(fix1,tx);
 866             fiy1             = _mm256_add_pd(fiy1,ty);
 867             fiz1             = _mm256_add_pd(fiz1,tz);
 868
 869             fjx0             = _mm256_add_pd(fjx0,tx);
 870             fjy0             = _mm256_add_pd(fjy0,ty);
 871             fjz0             = _mm256_add_pd(fjz0,tz);
 872
 873             /**************************
 874              * CALCULATE INTERACTIONS *
 875              **************************/
 876
 877             r20              = _mm256_mul_pd(rsq20,rinv20);
 878
 879             /* Compute parameters for interactions between i and j atoms */
 880             qq20             = _mm256_mul_pd(iq2,jq0);
 881
 882             /* Calculate table index by multiplying r with table scale and truncate to integer */
 883             rt               = _mm256_mul_pd(r20,vftabscale);
 884             vfitab           = _mm256_cvttpd_epi32(rt);
 885             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 886             vfitab           = _mm_slli_epi32(vfitab,2);
 887
 888             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 889             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 890             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 891             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 892             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 893             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 894             Heps             = _mm256_mul_pd(vfeps,H);
 895             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 896             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 897             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 898
 899             fscal            = felec;
 900
 901             /* Calculate temporary vectorial force */
 902             tx               = _mm256_mul_pd(fscal,dx20);
 903             ty               = _mm256_mul_pd(fscal,dy20);
 904             tz               = _mm256_mul_pd(fscal,dz20);
 905
 906             /* Update vectorial force */
 907             fix2             = _mm256_add_pd(fix2,tx);
 908             fiy2             = _mm256_add_pd(fiy2,ty);
 909             fiz2             = _mm256_add_pd(fiz2,tz);
 910
 911             fjx0             = _mm256_add_pd(fjx0,tx);
 912             fjy0             = _mm256_add_pd(fjy0,ty);
 913             fjz0             = _mm256_add_pd(fjz0,tz);
 914
 915             fjptrA             = f+j_coord_offsetA;
 916             fjptrB             = f+j_coord_offsetB;
 917             fjptrC             = f+j_coord_offsetC;
 918             fjptrD             = f+j_coord_offsetD;
 919
 920             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 921
 922             /* Inner loop uses 120 flops */
 923         }
 924
 925         if(jidx<j_index_end)
 926         {
 927
 928             /* Get j neighbor index, and coordinate index */
 929             jnrlistA         = jjnr[jidx];
 930             jnrlistB         = jjnr[jidx+1];
 931             jnrlistC         = jjnr[jidx+2];
 932             jnrlistD         = jjnr[jidx+3];
 933             /* Sign of each element will be negative for non-real atoms.
 934              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 935              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 936              */
 937             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 938
 939             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 940             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 941             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 942
 943             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 944             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 945             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 946             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 947             j_coord_offsetA  = DIM*jnrA;
 948             j_coord_offsetB  = DIM*jnrB;
 949             j_coord_offsetC  = DIM*jnrC;
 950             j_coord_offsetD  = DIM*jnrD;
 951
 952             /* load j atom coordinates */
 953             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 954                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 955                                                  &jx0,&jy0,&jz0);
 956
 957             /* Calculate displacement vector */
 958             dx00             = _mm256_sub_pd(ix0,jx0);
 959             dy00             = _mm256_sub_pd(iy0,jy0);
 960             dz00             = _mm256_sub_pd(iz0,jz0);
 961             dx10             = _mm256_sub_pd(ix1,jx0);
 962             dy10             = _mm256_sub_pd(iy1,jy0);
 963             dz10             = _mm256_sub_pd(iz1,jz0);
 964             dx20             = _mm256_sub_pd(ix2,jx0);
 965             dy20             = _mm256_sub_pd(iy2,jy0);
 966             dz20             = _mm256_sub_pd(iz2,jz0);
 967
 968             /* Calculate squared distance and things based on it */
 969             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 970             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 971             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 972
 973             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 974             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 975             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 976
 977             /* Load parameters for j particles */
 978             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 979                                                                  charge+jnrC+0,charge+jnrD+0);
 980
 981             fjx0             = _mm256_setzero_pd();
 982             fjy0             = _mm256_setzero_pd();
 983             fjz0             = _mm256_setzero_pd();
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             r00              = _mm256_mul_pd(rsq00,rinv00);
 990             r00              = _mm256_andnot_pd(dummy_mask,r00);
 991
 992             /* Compute parameters for interactions between i and j atoms */
 993             qq00             = _mm256_mul_pd(iq0,jq0);
 994
 995             /* Calculate table index by multiplying r with table scale and truncate to integer */
 996             rt               = _mm256_mul_pd(r00,vftabscale);
 997             vfitab           = _mm256_cvttpd_epi32(rt);
 998             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 999             vfitab           = _mm_slli_epi32(vfitab,2);
1000
1001             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1002             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1003             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1004             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1005             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1006             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1007             Heps             = _mm256_mul_pd(vfeps,H);
1008             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1009             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1010             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1011
1012             fscal            = felec;
1013
1014             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1015
1016             /* Calculate temporary vectorial force */
1017             tx               = _mm256_mul_pd(fscal,dx00);
1018             ty               = _mm256_mul_pd(fscal,dy00);
1019             tz               = _mm256_mul_pd(fscal,dz00);
1020
1021             /* Update vectorial force */
1022             fix0             = _mm256_add_pd(fix0,tx);
1023             fiy0             = _mm256_add_pd(fiy0,ty);
1024             fiz0             = _mm256_add_pd(fiz0,tz);
1025
1026             fjx0             = _mm256_add_pd(fjx0,tx);
1027             fjy0             = _mm256_add_pd(fjy0,ty);
1028             fjz0             = _mm256_add_pd(fjz0,tz);
1029
1030             /**************************
1031              * CALCULATE INTERACTIONS *
1032              **************************/
1033
1034             r10              = _mm256_mul_pd(rsq10,rinv10);
1035             r10              = _mm256_andnot_pd(dummy_mask,r10);
1036
1037             /* Compute parameters for interactions between i and j atoms */
1038             qq10             = _mm256_mul_pd(iq1,jq0);
1039
1040             /* Calculate table index by multiplying r with table scale and truncate to integer */
1041             rt               = _mm256_mul_pd(r10,vftabscale);
1042             vfitab           = _mm256_cvttpd_epi32(rt);
1043             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1044             vfitab           = _mm_slli_epi32(vfitab,2);
1045
1046             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1047             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1048             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1049             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1050             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1051             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1052             Heps             = _mm256_mul_pd(vfeps,H);
1053             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1054             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1055             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1056
1057             fscal            = felec;
1058
1059             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1060
1061             /* Calculate temporary vectorial force */
1062             tx               = _mm256_mul_pd(fscal,dx10);
1063             ty               = _mm256_mul_pd(fscal,dy10);
1064             tz               = _mm256_mul_pd(fscal,dz10);
1065
1066             /* Update vectorial force */
1067             fix1             = _mm256_add_pd(fix1,tx);
1068             fiy1             = _mm256_add_pd(fiy1,ty);
1069             fiz1             = _mm256_add_pd(fiz1,tz);
1070
1071             fjx0             = _mm256_add_pd(fjx0,tx);
1072             fjy0             = _mm256_add_pd(fjy0,ty);
1073             fjz0             = _mm256_add_pd(fjz0,tz);
1074
1075             /**************************
1076              * CALCULATE INTERACTIONS *
1077              **************************/
1078
1079             r20              = _mm256_mul_pd(rsq20,rinv20);
1080             r20              = _mm256_andnot_pd(dummy_mask,r20);
1081
1082             /* Compute parameters for interactions between i and j atoms */
1083             qq20             = _mm256_mul_pd(iq2,jq0);
1084
1085             /* Calculate table index by multiplying r with table scale and truncate to integer */
1086             rt               = _mm256_mul_pd(r20,vftabscale);
1087             vfitab           = _mm256_cvttpd_epi32(rt);
1088             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1089             vfitab           = _mm_slli_epi32(vfitab,2);
1090
1091             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1092             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1093             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1094             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1095             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1096             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1097             Heps             = _mm256_mul_pd(vfeps,H);
1098             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1099             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1100             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1101
1102             fscal            = felec;
1103
1104             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1105
1106             /* Calculate temporary vectorial force */
1107             tx               = _mm256_mul_pd(fscal,dx20);
1108             ty               = _mm256_mul_pd(fscal,dy20);
1109             tz               = _mm256_mul_pd(fscal,dz20);
1110
1111             /* Update vectorial force */
1112             fix2             = _mm256_add_pd(fix2,tx);
1113             fiy2             = _mm256_add_pd(fiy2,ty);
1114             fiz2             = _mm256_add_pd(fiz2,tz);
1115
1116             fjx0             = _mm256_add_pd(fjx0,tx);
1117             fjy0             = _mm256_add_pd(fjy0,ty);
1118             fjz0             = _mm256_add_pd(fjz0,tz);
1119
1120             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1121             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1122             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1123             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1124
1125             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1126
1127             /* Inner loop uses 123 flops */
1128         }
1129
1130         /* End of innermost loop */
1131
1132         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1133                                                  f+i_coord_offset,fshift+i_shift_offset);
1134
1135         /* Increment number of inner iterations */
1136         inneriter                  += j_index_end - j_index_start;
1137
1138         /* Outer loop uses 18 flops */
1139     }
1140
1141     /* Increment number of outer iterations */
1142     outeriter        += nri;
1143
1144     /* Update outer/inner flops */
1145
1146     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*123);
1147 }