src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwNone_GeomW4P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr1;
  85     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     real *           vdwioffsetptr2;
  87     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     real *           vdwioffsetptr3;
  89     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  95     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     __m128i          vfitab;
  98     __m128i          ifour       = _mm_set1_epi32(4);
  99     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 100     real             *vftab;
 101     __m256d          dummy_mask,cutoff_mask;
 102     __m128           tmpmask0,tmpmask1;
 103     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 104     __m256d          one     = _mm256_set1_pd(1.0);
 105     __m256d          two     = _mm256_set1_pd(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm256_set1_pd(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119
 120     vftab            = kernel_data->table_elec->data;
 121     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 126     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 127     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 160                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 161
 162         fix1             = _mm256_setzero_pd();
 163         fiy1             = _mm256_setzero_pd();
 164         fiz1             = _mm256_setzero_pd();
 165         fix2             = _mm256_setzero_pd();
 166         fiy2             = _mm256_setzero_pd();
 167         fiz2             = _mm256_setzero_pd();
 168         fix3             = _mm256_setzero_pd();
 169         fiy3             = _mm256_setzero_pd();
 170         fiz3             = _mm256_setzero_pd();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm256_setzero_pd();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186             j_coord_offsetC  = DIM*jnrC;
 187             j_coord_offsetD  = DIM*jnrD;
 188
 189             /* load j atom coordinates */
 190             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 192                                                  &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx10             = _mm256_sub_pd(ix1,jx0);
 196             dy10             = _mm256_sub_pd(iy1,jy0);
 197             dz10             = _mm256_sub_pd(iz1,jz0);
 198             dx20             = _mm256_sub_pd(ix2,jx0);
 199             dy20             = _mm256_sub_pd(iy2,jy0);
 200             dz20             = _mm256_sub_pd(iz2,jz0);
 201             dx30             = _mm256_sub_pd(ix3,jx0);
 202             dy30             = _mm256_sub_pd(iy3,jy0);
 203             dz30             = _mm256_sub_pd(iz3,jz0);
 204
 205             /* Calculate squared distance and things based on it */
 206             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 207             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 208             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 209
 210             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 211             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 212             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 216                                                                  charge+jnrC+0,charge+jnrD+0);
 217
 218             fjx0             = _mm256_setzero_pd();
 219             fjy0             = _mm256_setzero_pd();
 220             fjz0             = _mm256_setzero_pd();
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             r10              = _mm256_mul_pd(rsq10,rinv10);
 227
 228             /* Compute parameters for interactions between i and j atoms */
 229             qq10             = _mm256_mul_pd(iq1,jq0);
 230
 231             /* Calculate table index by multiplying r with table scale and truncate to integer */
 232             rt               = _mm256_mul_pd(r10,vftabscale);
 233             vfitab           = _mm256_cvttpd_epi32(rt);
 234             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 235             vfitab           = _mm_slli_epi32(vfitab,2);
 236
 237             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 238             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 239             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 240             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 241             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 242             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 243             Heps             = _mm256_mul_pd(vfeps,H);
 244             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 245             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 246             velec            = _mm256_mul_pd(qq10,VV);
 247             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 248             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 249
 250             /* Update potential sum for this i atom from the interaction with this j atom. */
 251             velecsum         = _mm256_add_pd(velecsum,velec);
 252
 253             fscal            = felec;
 254
 255             /* Calculate temporary vectorial force */
 256             tx               = _mm256_mul_pd(fscal,dx10);
 257             ty               = _mm256_mul_pd(fscal,dy10);
 258             tz               = _mm256_mul_pd(fscal,dz10);
 259
 260             /* Update vectorial force */
 261             fix1             = _mm256_add_pd(fix1,tx);
 262             fiy1             = _mm256_add_pd(fiy1,ty);
 263             fiz1             = _mm256_add_pd(fiz1,tz);
 264
 265             fjx0             = _mm256_add_pd(fjx0,tx);
 266             fjy0             = _mm256_add_pd(fjy0,ty);
 267             fjz0             = _mm256_add_pd(fjz0,tz);
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             r20              = _mm256_mul_pd(rsq20,rinv20);
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             qq20             = _mm256_mul_pd(iq2,jq0);
 277
 278             /* Calculate table index by multiplying r with table scale and truncate to integer */
 279             rt               = _mm256_mul_pd(r20,vftabscale);
 280             vfitab           = _mm256_cvttpd_epi32(rt);
 281             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 282             vfitab           = _mm_slli_epi32(vfitab,2);
 283
 284             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 285             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 286             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 287             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 288             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 289             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 290             Heps             = _mm256_mul_pd(vfeps,H);
 291             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 292             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 293             velec            = _mm256_mul_pd(qq20,VV);
 294             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 295             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm256_add_pd(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm256_mul_pd(fscal,dx20);
 304             ty               = _mm256_mul_pd(fscal,dy20);
 305             tz               = _mm256_mul_pd(fscal,dz20);
 306
 307             /* Update vectorial force */
 308             fix2             = _mm256_add_pd(fix2,tx);
 309             fiy2             = _mm256_add_pd(fiy2,ty);
 310             fiz2             = _mm256_add_pd(fiz2,tz);
 311
 312             fjx0             = _mm256_add_pd(fjx0,tx);
 313             fjy0             = _mm256_add_pd(fjy0,ty);
 314             fjz0             = _mm256_add_pd(fjz0,tz);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             r30              = _mm256_mul_pd(rsq30,rinv30);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq30             = _mm256_mul_pd(iq3,jq0);
 324
 325             /* Calculate table index by multiplying r with table scale and truncate to integer */
 326             rt               = _mm256_mul_pd(r30,vftabscale);
 327             vfitab           = _mm256_cvttpd_epi32(rt);
 328             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 329             vfitab           = _mm_slli_epi32(vfitab,2);
 330
 331             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 332             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 333             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 334             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 335             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 336             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 337             Heps             = _mm256_mul_pd(vfeps,H);
 338             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 339             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 340             velec            = _mm256_mul_pd(qq30,VV);
 341             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 342             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm256_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_pd(fscal,dx30);
 351             ty               = _mm256_mul_pd(fscal,dy30);
 352             tz               = _mm256_mul_pd(fscal,dz30);
 353
 354             /* Update vectorial force */
 355             fix3             = _mm256_add_pd(fix3,tx);
 356             fiy3             = _mm256_add_pd(fiy3,ty);
 357             fiz3             = _mm256_add_pd(fiz3,tz);
 358
 359             fjx0             = _mm256_add_pd(fjx0,tx);
 360             fjy0             = _mm256_add_pd(fjy0,ty);
 361             fjz0             = _mm256_add_pd(fjz0,tz);
 362
 363             fjptrA             = f+j_coord_offsetA;
 364             fjptrB             = f+j_coord_offsetB;
 365             fjptrC             = f+j_coord_offsetC;
 366             fjptrD             = f+j_coord_offsetD;
 367
 368             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 369
 370             /* Inner loop uses 132 flops */
 371         }
 372
 373         if(jidx<j_index_end)
 374         {
 375
 376             /* Get j neighbor index, and coordinate index */
 377             jnrlistA         = jjnr[jidx];
 378             jnrlistB         = jjnr[jidx+1];
 379             jnrlistC         = jjnr[jidx+2];
 380             jnrlistD         = jjnr[jidx+3];
 381             /* Sign of each element will be negative for non-real atoms.
 382              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 383              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 384              */
 385             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 386
 387             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 388             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 389             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 390
 391             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 392             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 393             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 394             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 395             j_coord_offsetA  = DIM*jnrA;
 396             j_coord_offsetB  = DIM*jnrB;
 397             j_coord_offsetC  = DIM*jnrC;
 398             j_coord_offsetD  = DIM*jnrD;
 399
 400             /* load j atom coordinates */
 401             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 402                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 403                                                  &jx0,&jy0,&jz0);
 404
 405             /* Calculate displacement vector */
 406             dx10             = _mm256_sub_pd(ix1,jx0);
 407             dy10             = _mm256_sub_pd(iy1,jy0);
 408             dz10             = _mm256_sub_pd(iz1,jz0);
 409             dx20             = _mm256_sub_pd(ix2,jx0);
 410             dy20             = _mm256_sub_pd(iy2,jy0);
 411             dz20             = _mm256_sub_pd(iz2,jz0);
 412             dx30             = _mm256_sub_pd(ix3,jx0);
 413             dy30             = _mm256_sub_pd(iy3,jy0);
 414             dz30             = _mm256_sub_pd(iz3,jz0);
 415
 416             /* Calculate squared distance and things based on it */
 417             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 418             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 419             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 420
 421             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 422             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 423             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 424
 425             /* Load parameters for j particles */
 426             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 427                                                                  charge+jnrC+0,charge+jnrD+0);
 428
 429             fjx0             = _mm256_setzero_pd();
 430             fjy0             = _mm256_setzero_pd();
 431             fjz0             = _mm256_setzero_pd();
 432
 433             /**************************
 434              * CALCULATE INTERACTIONS *
 435              **************************/
 436
 437             r10              = _mm256_mul_pd(rsq10,rinv10);
 438             r10              = _mm256_andnot_pd(dummy_mask,r10);
 439
 440             /* Compute parameters for interactions between i and j atoms */
 441             qq10             = _mm256_mul_pd(iq1,jq0);
 442
 443             /* Calculate table index by multiplying r with table scale and truncate to integer */
 444             rt               = _mm256_mul_pd(r10,vftabscale);
 445             vfitab           = _mm256_cvttpd_epi32(rt);
 446             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 447             vfitab           = _mm_slli_epi32(vfitab,2);
 448
 449             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 450             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 451             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 452             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 453             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 454             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 455             Heps             = _mm256_mul_pd(vfeps,H);
 456             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 457             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 458             velec            = _mm256_mul_pd(qq10,VV);
 459             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 460             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 461
 462             /* Update potential sum for this i atom from the interaction with this j atom. */
 463             velec            = _mm256_andnot_pd(dummy_mask,velec);
 464             velecsum         = _mm256_add_pd(velecsum,velec);
 465
 466             fscal            = felec;
 467
 468             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 469
 470             /* Calculate temporary vectorial force */
 471             tx               = _mm256_mul_pd(fscal,dx10);
 472             ty               = _mm256_mul_pd(fscal,dy10);
 473             tz               = _mm256_mul_pd(fscal,dz10);
 474
 475             /* Update vectorial force */
 476             fix1             = _mm256_add_pd(fix1,tx);
 477             fiy1             = _mm256_add_pd(fiy1,ty);
 478             fiz1             = _mm256_add_pd(fiz1,tz);
 479
 480             fjx0             = _mm256_add_pd(fjx0,tx);
 481             fjy0             = _mm256_add_pd(fjy0,ty);
 482             fjz0             = _mm256_add_pd(fjz0,tz);
 483
 484             /**************************
 485              * CALCULATE INTERACTIONS *
 486              **************************/
 487
 488             r20              = _mm256_mul_pd(rsq20,rinv20);
 489             r20              = _mm256_andnot_pd(dummy_mask,r20);
 490
 491             /* Compute parameters for interactions between i and j atoms */
 492             qq20             = _mm256_mul_pd(iq2,jq0);
 493
 494             /* Calculate table index by multiplying r with table scale and truncate to integer */
 495             rt               = _mm256_mul_pd(r20,vftabscale);
 496             vfitab           = _mm256_cvttpd_epi32(rt);
 497             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 498             vfitab           = _mm_slli_epi32(vfitab,2);
 499
 500             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 501             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 502             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 503             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 504             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 505             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 506             Heps             = _mm256_mul_pd(vfeps,H);
 507             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 508             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 509             velec            = _mm256_mul_pd(qq20,VV);
 510             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 511             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             velec            = _mm256_andnot_pd(dummy_mask,velec);
 515             velecsum         = _mm256_add_pd(velecsum,velec);
 516
 517             fscal            = felec;
 518
 519             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm256_mul_pd(fscal,dx20);
 523             ty               = _mm256_mul_pd(fscal,dy20);
 524             tz               = _mm256_mul_pd(fscal,dz20);
 525
 526             /* Update vectorial force */
 527             fix2             = _mm256_add_pd(fix2,tx);
 528             fiy2             = _mm256_add_pd(fiy2,ty);
 529             fiz2             = _mm256_add_pd(fiz2,tz);
 530
 531             fjx0             = _mm256_add_pd(fjx0,tx);
 532             fjy0             = _mm256_add_pd(fjy0,ty);
 533             fjz0             = _mm256_add_pd(fjz0,tz);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             r30              = _mm256_mul_pd(rsq30,rinv30);
 540             r30              = _mm256_andnot_pd(dummy_mask,r30);
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             qq30             = _mm256_mul_pd(iq3,jq0);
 544
 545             /* Calculate table index by multiplying r with table scale and truncate to integer */
 546             rt               = _mm256_mul_pd(r30,vftabscale);
 547             vfitab           = _mm256_cvttpd_epi32(rt);
 548             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 549             vfitab           = _mm_slli_epi32(vfitab,2);
 550
 551             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 552             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 553             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 554             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 555             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 556             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 557             Heps             = _mm256_mul_pd(vfeps,H);
 558             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 559             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 560             velec            = _mm256_mul_pd(qq30,VV);
 561             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 562             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             velec            = _mm256_andnot_pd(dummy_mask,velec);
 566             velecsum         = _mm256_add_pd(velecsum,velec);
 567
 568             fscal            = felec;
 569
 570             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 571
 572             /* Calculate temporary vectorial force */
 573             tx               = _mm256_mul_pd(fscal,dx30);
 574             ty               = _mm256_mul_pd(fscal,dy30);
 575             tz               = _mm256_mul_pd(fscal,dz30);
 576
 577             /* Update vectorial force */
 578             fix3             = _mm256_add_pd(fix3,tx);
 579             fiy3             = _mm256_add_pd(fiy3,ty);
 580             fiz3             = _mm256_add_pd(fiz3,tz);
 581
 582             fjx0             = _mm256_add_pd(fjx0,tx);
 583             fjy0             = _mm256_add_pd(fjy0,ty);
 584             fjz0             = _mm256_add_pd(fjz0,tz);
 585
 586             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 587             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 588             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 589             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 590
 591             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 592
 593             /* Inner loop uses 135 flops */
 594         }
 595
 596         /* End of innermost loop */
 597
 598         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 599                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 600
 601         ggid                        = gid[iidx];
 602         /* Update potential energies */
 603         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 604
 605         /* Increment number of inner iterations */
 606         inneriter                  += j_index_end - j_index_start;
 607
 608         /* Outer loop uses 19 flops */
 609     }
 610
 611     /* Increment number of outer iterations */
 612     outeriter        += nri;
 613
 614     /* Update outer/inner flops */
 615
 616     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*135);
 617 }
 618 /*
 619  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_256_double
 620  * Electrostatics interaction: CubicSplineTable
 621  * VdW interaction:            None
 622  * Geometry:                   Water4-Particle
 623  * Calculate force/pot:        Force
 624  */
 625 void
 626 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_256_double
 627                     (t_nblist                    * gmx_restrict       nlist,
 628                      rvec                        * gmx_restrict          xx,
 629                      rvec                        * gmx_restrict          ff,
 630                      t_forcerec                  * gmx_restrict          fr,
 631                      t_mdatoms                   * gmx_restrict     mdatoms,
 632                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 633                      t_nrnb                      * gmx_restrict        nrnb)
 634 {
 635     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 636      * just 0 for non-waters.
 637      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 638      * jnr indices corresponding to data put in the four positions in the SIMD register.
 639      */
 640     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 641     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 642     int              jnrA,jnrB,jnrC,jnrD;
 643     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 644     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 645     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 646     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 647     real             rcutoff_scalar;
 648     real             *shiftvec,*fshift,*x,*f;
 649     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 650     real             scratch[4*DIM];
 651     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 652     real *           vdwioffsetptr1;
 653     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 654     real *           vdwioffsetptr2;
 655     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 656     real *           vdwioffsetptr3;
 657     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 658     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 659     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 660     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 661     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 662     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 663     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 664     real             *charge;
 665     __m128i          vfitab;
 666     __m128i          ifour       = _mm_set1_epi32(4);
 667     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 668     real             *vftab;
 669     __m256d          dummy_mask,cutoff_mask;
 670     __m128           tmpmask0,tmpmask1;
 671     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 672     __m256d          one     = _mm256_set1_pd(1.0);
 673     __m256d          two     = _mm256_set1_pd(2.0);
 674     x                = xx[0];
 675     f                = ff[0];
 676
 677     nri              = nlist->nri;
 678     iinr             = nlist->iinr;
 679     jindex           = nlist->jindex;
 680     jjnr             = nlist->jjnr;
 681     shiftidx         = nlist->shift;
 682     gid              = nlist->gid;
 683     shiftvec         = fr->shift_vec[0];
 684     fshift           = fr->fshift[0];
 685     facel            = _mm256_set1_pd(fr->epsfac);
 686     charge           = mdatoms->chargeA;
 687
 688     vftab            = kernel_data->table_elec->data;
 689     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 690
 691     /* Setup water-specific parameters */
 692     inr              = nlist->iinr[0];
 693     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 694     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 695     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 696
 697     /* Avoid stupid compiler warnings */
 698     jnrA = jnrB = jnrC = jnrD = 0;
 699     j_coord_offsetA = 0;
 700     j_coord_offsetB = 0;
 701     j_coord_offsetC = 0;
 702     j_coord_offsetD = 0;
 703
 704     outeriter        = 0;
 705     inneriter        = 0;
 706
 707     for(iidx=0;iidx<4*DIM;iidx++)
 708     {
 709         scratch[iidx] = 0.0;
 710     }
 711
 712     /* Start outer loop over neighborlists */
 713     for(iidx=0; iidx<nri; iidx++)
 714     {
 715         /* Load shift vector for this list */
 716         i_shift_offset   = DIM*shiftidx[iidx];
 717
 718         /* Load limits for loop over neighbors */
 719         j_index_start    = jindex[iidx];
 720         j_index_end      = jindex[iidx+1];
 721
 722         /* Get outer coordinate index */
 723         inr              = iinr[iidx];
 724         i_coord_offset   = DIM*inr;
 725
 726         /* Load i particle coords and add shift vector */
 727         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 728                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 729
 730         fix1             = _mm256_setzero_pd();
 731         fiy1             = _mm256_setzero_pd();
 732         fiz1             = _mm256_setzero_pd();
 733         fix2             = _mm256_setzero_pd();
 734         fiy2             = _mm256_setzero_pd();
 735         fiz2             = _mm256_setzero_pd();
 736         fix3             = _mm256_setzero_pd();
 737         fiy3             = _mm256_setzero_pd();
 738         fiz3             = _mm256_setzero_pd();
 739
 740         /* Start inner kernel loop */
 741         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 742         {
 743
 744             /* Get j neighbor index, and coordinate index */
 745             jnrA             = jjnr[jidx];
 746             jnrB             = jjnr[jidx+1];
 747             jnrC             = jjnr[jidx+2];
 748             jnrD             = jjnr[jidx+3];
 749             j_coord_offsetA  = DIM*jnrA;
 750             j_coord_offsetB  = DIM*jnrB;
 751             j_coord_offsetC  = DIM*jnrC;
 752             j_coord_offsetD  = DIM*jnrD;
 753
 754             /* load j atom coordinates */
 755             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 756                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 757                                                  &jx0,&jy0,&jz0);
 758
 759             /* Calculate displacement vector */
 760             dx10             = _mm256_sub_pd(ix1,jx0);
 761             dy10             = _mm256_sub_pd(iy1,jy0);
 762             dz10             = _mm256_sub_pd(iz1,jz0);
 763             dx20             = _mm256_sub_pd(ix2,jx0);
 764             dy20             = _mm256_sub_pd(iy2,jy0);
 765             dz20             = _mm256_sub_pd(iz2,jz0);
 766             dx30             = _mm256_sub_pd(ix3,jx0);
 767             dy30             = _mm256_sub_pd(iy3,jy0);
 768             dz30             = _mm256_sub_pd(iz3,jz0);
 769
 770             /* Calculate squared distance and things based on it */
 771             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 772             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 773             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 774
 775             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 776             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 777             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 778
 779             /* Load parameters for j particles */
 780             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 781                                                                  charge+jnrC+0,charge+jnrD+0);
 782
 783             fjx0             = _mm256_setzero_pd();
 784             fjy0             = _mm256_setzero_pd();
 785             fjz0             = _mm256_setzero_pd();
 786
 787             /**************************
 788              * CALCULATE INTERACTIONS *
 789              **************************/
 790
 791             r10              = _mm256_mul_pd(rsq10,rinv10);
 792
 793             /* Compute parameters for interactions between i and j atoms */
 794             qq10             = _mm256_mul_pd(iq1,jq0);
 795
 796             /* Calculate table index by multiplying r with table scale and truncate to integer */
 797             rt               = _mm256_mul_pd(r10,vftabscale);
 798             vfitab           = _mm256_cvttpd_epi32(rt);
 799             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 800             vfitab           = _mm_slli_epi32(vfitab,2);
 801
 802             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 803             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 804             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 805             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 806             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 807             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 808             Heps             = _mm256_mul_pd(vfeps,H);
 809             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 810             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 811             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 812
 813             fscal            = felec;
 814
 815             /* Calculate temporary vectorial force */
 816             tx               = _mm256_mul_pd(fscal,dx10);
 817             ty               = _mm256_mul_pd(fscal,dy10);
 818             tz               = _mm256_mul_pd(fscal,dz10);
 819
 820             /* Update vectorial force */
 821             fix1             = _mm256_add_pd(fix1,tx);
 822             fiy1             = _mm256_add_pd(fiy1,ty);
 823             fiz1             = _mm256_add_pd(fiz1,tz);
 824
 825             fjx0             = _mm256_add_pd(fjx0,tx);
 826             fjy0             = _mm256_add_pd(fjy0,ty);
 827             fjz0             = _mm256_add_pd(fjz0,tz);
 828
 829             /**************************
 830              * CALCULATE INTERACTIONS *
 831              **************************/
 832
 833             r20              = _mm256_mul_pd(rsq20,rinv20);
 834
 835             /* Compute parameters for interactions between i and j atoms */
 836             qq20             = _mm256_mul_pd(iq2,jq0);
 837
 838             /* Calculate table index by multiplying r with table scale and truncate to integer */
 839             rt               = _mm256_mul_pd(r20,vftabscale);
 840             vfitab           = _mm256_cvttpd_epi32(rt);
 841             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 842             vfitab           = _mm_slli_epi32(vfitab,2);
 843
 844             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 845             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 846             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 847             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 848             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 849             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 850             Heps             = _mm256_mul_pd(vfeps,H);
 851             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 852             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 853             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 854
 855             fscal            = felec;
 856
 857             /* Calculate temporary vectorial force */
 858             tx               = _mm256_mul_pd(fscal,dx20);
 859             ty               = _mm256_mul_pd(fscal,dy20);
 860             tz               = _mm256_mul_pd(fscal,dz20);
 861
 862             /* Update vectorial force */
 863             fix2             = _mm256_add_pd(fix2,tx);
 864             fiy2             = _mm256_add_pd(fiy2,ty);
 865             fiz2             = _mm256_add_pd(fiz2,tz);
 866
 867             fjx0             = _mm256_add_pd(fjx0,tx);
 868             fjy0             = _mm256_add_pd(fjy0,ty);
 869             fjz0             = _mm256_add_pd(fjz0,tz);
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             r30              = _mm256_mul_pd(rsq30,rinv30);
 876
 877             /* Compute parameters for interactions between i and j atoms */
 878             qq30             = _mm256_mul_pd(iq3,jq0);
 879
 880             /* Calculate table index by multiplying r with table scale and truncate to integer */
 881             rt               = _mm256_mul_pd(r30,vftabscale);
 882             vfitab           = _mm256_cvttpd_epi32(rt);
 883             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 884             vfitab           = _mm_slli_epi32(vfitab,2);
 885
 886             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 887             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 888             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 889             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 890             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 891             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 892             Heps             = _mm256_mul_pd(vfeps,H);
 893             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 894             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 895             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
 896
 897             fscal            = felec;
 898
 899             /* Calculate temporary vectorial force */
 900             tx               = _mm256_mul_pd(fscal,dx30);
 901             ty               = _mm256_mul_pd(fscal,dy30);
 902             tz               = _mm256_mul_pd(fscal,dz30);
 903
 904             /* Update vectorial force */
 905             fix3             = _mm256_add_pd(fix3,tx);
 906             fiy3             = _mm256_add_pd(fiy3,ty);
 907             fiz3             = _mm256_add_pd(fiz3,tz);
 908
 909             fjx0             = _mm256_add_pd(fjx0,tx);
 910             fjy0             = _mm256_add_pd(fjy0,ty);
 911             fjz0             = _mm256_add_pd(fjz0,tz);
 912
 913             fjptrA             = f+j_coord_offsetA;
 914             fjptrB             = f+j_coord_offsetB;
 915             fjptrC             = f+j_coord_offsetC;
 916             fjptrD             = f+j_coord_offsetD;
 917
 918             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 919
 920             /* Inner loop uses 120 flops */
 921         }
 922
 923         if(jidx<j_index_end)
 924         {
 925
 926             /* Get j neighbor index, and coordinate index */
 927             jnrlistA         = jjnr[jidx];
 928             jnrlistB         = jjnr[jidx+1];
 929             jnrlistC         = jjnr[jidx+2];
 930             jnrlistD         = jjnr[jidx+3];
 931             /* Sign of each element will be negative for non-real atoms.
 932              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 933              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 934              */
 935             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 936
 937             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 938             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 939             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 940
 941             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 942             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 943             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 944             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 945             j_coord_offsetA  = DIM*jnrA;
 946             j_coord_offsetB  = DIM*jnrB;
 947             j_coord_offsetC  = DIM*jnrC;
 948             j_coord_offsetD  = DIM*jnrD;
 949
 950             /* load j atom coordinates */
 951             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 952                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 953                                                  &jx0,&jy0,&jz0);
 954
 955             /* Calculate displacement vector */
 956             dx10             = _mm256_sub_pd(ix1,jx0);
 957             dy10             = _mm256_sub_pd(iy1,jy0);
 958             dz10             = _mm256_sub_pd(iz1,jz0);
 959             dx20             = _mm256_sub_pd(ix2,jx0);
 960             dy20             = _mm256_sub_pd(iy2,jy0);
 961             dz20             = _mm256_sub_pd(iz2,jz0);
 962             dx30             = _mm256_sub_pd(ix3,jx0);
 963             dy30             = _mm256_sub_pd(iy3,jy0);
 964             dz30             = _mm256_sub_pd(iz3,jz0);
 965
 966             /* Calculate squared distance and things based on it */
 967             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 968             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 969             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 970
 971             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 972             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 973             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 974
 975             /* Load parameters for j particles */
 976             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 977                                                                  charge+jnrC+0,charge+jnrD+0);
 978
 979             fjx0             = _mm256_setzero_pd();
 980             fjy0             = _mm256_setzero_pd();
 981             fjz0             = _mm256_setzero_pd();
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             r10              = _mm256_mul_pd(rsq10,rinv10);
 988             r10              = _mm256_andnot_pd(dummy_mask,r10);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq10             = _mm256_mul_pd(iq1,jq0);
 992
 993             /* Calculate table index by multiplying r with table scale and truncate to integer */
 994             rt               = _mm256_mul_pd(r10,vftabscale);
 995             vfitab           = _mm256_cvttpd_epi32(rt);
 996             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 997             vfitab           = _mm_slli_epi32(vfitab,2);
 998
 999             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1000             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1001             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1002             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1003             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1004             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1005             Heps             = _mm256_mul_pd(vfeps,H);
1006             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1007             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1008             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1009
1010             fscal            = felec;
1011
1012             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm256_mul_pd(fscal,dx10);
1016             ty               = _mm256_mul_pd(fscal,dy10);
1017             tz               = _mm256_mul_pd(fscal,dz10);
1018
1019             /* Update vectorial force */
1020             fix1             = _mm256_add_pd(fix1,tx);
1021             fiy1             = _mm256_add_pd(fiy1,ty);
1022             fiz1             = _mm256_add_pd(fiz1,tz);
1023
1024             fjx0             = _mm256_add_pd(fjx0,tx);
1025             fjy0             = _mm256_add_pd(fjy0,ty);
1026             fjz0             = _mm256_add_pd(fjz0,tz);
1027
1028             /**************************
1029              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             r20              = _mm256_mul_pd(rsq20,rinv20);
1033             r20              = _mm256_andnot_pd(dummy_mask,r20);
1034
1035             /* Compute parameters for interactions between i and j atoms */
1036             qq20             = _mm256_mul_pd(iq2,jq0);
1037
1038             /* Calculate table index by multiplying r with table scale and truncate to integer */
1039             rt               = _mm256_mul_pd(r20,vftabscale);
1040             vfitab           = _mm256_cvttpd_epi32(rt);
1041             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1042             vfitab           = _mm_slli_epi32(vfitab,2);
1043
1044             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1045             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1046             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1047             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1048             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1049             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1050             Heps             = _mm256_mul_pd(vfeps,H);
1051             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1052             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1053             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1054
1055             fscal            = felec;
1056
1057             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1058
1059             /* Calculate temporary vectorial force */
1060             tx               = _mm256_mul_pd(fscal,dx20);
1061             ty               = _mm256_mul_pd(fscal,dy20);
1062             tz               = _mm256_mul_pd(fscal,dz20);
1063
1064             /* Update vectorial force */
1065             fix2             = _mm256_add_pd(fix2,tx);
1066             fiy2             = _mm256_add_pd(fiy2,ty);
1067             fiz2             = _mm256_add_pd(fiz2,tz);
1068
1069             fjx0             = _mm256_add_pd(fjx0,tx);
1070             fjy0             = _mm256_add_pd(fjy0,ty);
1071             fjz0             = _mm256_add_pd(fjz0,tz);
1072
1073             /**************************
1074              * CALCULATE INTERACTIONS *
1075              **************************/
1076
1077             r30              = _mm256_mul_pd(rsq30,rinv30);
1078             r30              = _mm256_andnot_pd(dummy_mask,r30);
1079
1080             /* Compute parameters for interactions between i and j atoms */
1081             qq30             = _mm256_mul_pd(iq3,jq0);
1082
1083             /* Calculate table index by multiplying r with table scale and truncate to integer */
1084             rt               = _mm256_mul_pd(r30,vftabscale);
1085             vfitab           = _mm256_cvttpd_epi32(rt);
1086             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1087             vfitab           = _mm_slli_epi32(vfitab,2);
1088
1089             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1090             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1091             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1092             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1093             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1094             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1095             Heps             = _mm256_mul_pd(vfeps,H);
1096             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1097             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1098             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
1099
1100             fscal            = felec;
1101
1102             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1103
1104             /* Calculate temporary vectorial force */
1105             tx               = _mm256_mul_pd(fscal,dx30);
1106             ty               = _mm256_mul_pd(fscal,dy30);
1107             tz               = _mm256_mul_pd(fscal,dz30);
1108
1109             /* Update vectorial force */
1110             fix3             = _mm256_add_pd(fix3,tx);
1111             fiy3             = _mm256_add_pd(fiy3,ty);
1112             fiz3             = _mm256_add_pd(fiz3,tz);
1113
1114             fjx0             = _mm256_add_pd(fjx0,tx);
1115             fjy0             = _mm256_add_pd(fjy0,ty);
1116             fjz0             = _mm256_add_pd(fjz0,tz);
1117
1118             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1119             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1120             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1121             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1122
1123             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1124
1125             /* Inner loop uses 123 flops */
1126         }
1127
1128         /* End of innermost loop */
1129
1130         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1131                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1132
1133         /* Increment number of inner iterations */
1134         inneriter                  += j_index_end - j_index_start;
1135
1136         /* Outer loop uses 18 flops */
1137     }
1138
1139     /* Increment number of outer iterations */
1140     outeriter        += nri;
1141
1142     /* Update outer/inner flops */
1143
1144     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*123);
1145 }