src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwNone_GeomW4P1_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 *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     __m256d          dummy_mask,cutoff_mask;
 100     __m128           tmpmask0,tmpmask1;
 101     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 102     __m256d          one     = _mm256_set1_pd(1.0);
 103     __m256d          two     = _mm256_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm256_set1_pd(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117
 118     /* Setup water-specific parameters */
 119     inr              = nlist->iinr[0];
 120     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 121     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 122     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = jnrC = jnrD = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128     j_coord_offsetC = 0;
 129     j_coord_offsetD = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     for(iidx=0;iidx<4*DIM;iidx++)
 135     {
 136         scratch[iidx] = 0.0;
 137     }
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 155                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 156
 157         fix1             = _mm256_setzero_pd();
 158         fiy1             = _mm256_setzero_pd();
 159         fiz1             = _mm256_setzero_pd();
 160         fix2             = _mm256_setzero_pd();
 161         fiy2             = _mm256_setzero_pd();
 162         fiz2             = _mm256_setzero_pd();
 163         fix3             = _mm256_setzero_pd();
 164         fiy3             = _mm256_setzero_pd();
 165         fiz3             = _mm256_setzero_pd();
 166
 167         /* Reset potential sums */
 168         velecsum         = _mm256_setzero_pd();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             j_coord_offsetA  = DIM*jnrA;
 180             j_coord_offsetB  = DIM*jnrB;
 181             j_coord_offsetC  = DIM*jnrC;
 182             j_coord_offsetD  = DIM*jnrD;
 183
 184             /* load j atom coordinates */
 185             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 187                                                  &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx10             = _mm256_sub_pd(ix1,jx0);
 191             dy10             = _mm256_sub_pd(iy1,jy0);
 192             dz10             = _mm256_sub_pd(iz1,jz0);
 193             dx20             = _mm256_sub_pd(ix2,jx0);
 194             dy20             = _mm256_sub_pd(iy2,jy0);
 195             dz20             = _mm256_sub_pd(iz2,jz0);
 196             dx30             = _mm256_sub_pd(ix3,jx0);
 197             dy30             = _mm256_sub_pd(iy3,jy0);
 198             dz30             = _mm256_sub_pd(iz3,jz0);
 199
 200             /* Calculate squared distance and things based on it */
 201             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 202             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 203             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 204
 205             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 206             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 207             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 208
 209             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 210             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 211             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 212
 213             /* Load parameters for j particles */
 214             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 215                                                                  charge+jnrC+0,charge+jnrD+0);
 216
 217             fjx0             = _mm256_setzero_pd();
 218             fjy0             = _mm256_setzero_pd();
 219             fjz0             = _mm256_setzero_pd();
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq10             = _mm256_mul_pd(iq1,jq0);
 227
 228             /* COULOMB ELECTROSTATICS */
 229             velec            = _mm256_mul_pd(qq10,rinv10);
 230             felec            = _mm256_mul_pd(velec,rinvsq10);
 231
 232             /* Update potential sum for this i atom from the interaction with this j atom. */
 233             velecsum         = _mm256_add_pd(velecsum,velec);
 234
 235             fscal            = felec;
 236
 237             /* Calculate temporary vectorial force */
 238             tx               = _mm256_mul_pd(fscal,dx10);
 239             ty               = _mm256_mul_pd(fscal,dy10);
 240             tz               = _mm256_mul_pd(fscal,dz10);
 241
 242             /* Update vectorial force */
 243             fix1             = _mm256_add_pd(fix1,tx);
 244             fiy1             = _mm256_add_pd(fiy1,ty);
 245             fiz1             = _mm256_add_pd(fiz1,tz);
 246
 247             fjx0             = _mm256_add_pd(fjx0,tx);
 248             fjy0             = _mm256_add_pd(fjy0,ty);
 249             fjz0             = _mm256_add_pd(fjz0,tz);
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             qq20             = _mm256_mul_pd(iq2,jq0);
 257
 258             /* COULOMB ELECTROSTATICS */
 259             velec            = _mm256_mul_pd(qq20,rinv20);
 260             felec            = _mm256_mul_pd(velec,rinvsq20);
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             velecsum         = _mm256_add_pd(velecsum,velec);
 264
 265             fscal            = felec;
 266
 267             /* Calculate temporary vectorial force */
 268             tx               = _mm256_mul_pd(fscal,dx20);
 269             ty               = _mm256_mul_pd(fscal,dy20);
 270             tz               = _mm256_mul_pd(fscal,dz20);
 271
 272             /* Update vectorial force */
 273             fix2             = _mm256_add_pd(fix2,tx);
 274             fiy2             = _mm256_add_pd(fiy2,ty);
 275             fiz2             = _mm256_add_pd(fiz2,tz);
 276
 277             fjx0             = _mm256_add_pd(fjx0,tx);
 278             fjy0             = _mm256_add_pd(fjy0,ty);
 279             fjz0             = _mm256_add_pd(fjz0,tz);
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq30             = _mm256_mul_pd(iq3,jq0);
 287
 288             /* COULOMB ELECTROSTATICS */
 289             velec            = _mm256_mul_pd(qq30,rinv30);
 290             felec            = _mm256_mul_pd(velec,rinvsq30);
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velecsum         = _mm256_add_pd(velecsum,velec);
 294
 295             fscal            = felec;
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm256_mul_pd(fscal,dx30);
 299             ty               = _mm256_mul_pd(fscal,dy30);
 300             tz               = _mm256_mul_pd(fscal,dz30);
 301
 302             /* Update vectorial force */
 303             fix3             = _mm256_add_pd(fix3,tx);
 304             fiy3             = _mm256_add_pd(fiy3,ty);
 305             fiz3             = _mm256_add_pd(fiz3,tz);
 306
 307             fjx0             = _mm256_add_pd(fjx0,tx);
 308             fjy0             = _mm256_add_pd(fjy0,ty);
 309             fjz0             = _mm256_add_pd(fjz0,tz);
 310
 311             fjptrA             = f+j_coord_offsetA;
 312             fjptrB             = f+j_coord_offsetB;
 313             fjptrC             = f+j_coord_offsetC;
 314             fjptrD             = f+j_coord_offsetD;
 315
 316             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 317
 318             /* Inner loop uses 84 flops */
 319         }
 320
 321         if(jidx<j_index_end)
 322         {
 323
 324             /* Get j neighbor index, and coordinate index */
 325             jnrlistA         = jjnr[jidx];
 326             jnrlistB         = jjnr[jidx+1];
 327             jnrlistC         = jjnr[jidx+2];
 328             jnrlistD         = jjnr[jidx+3];
 329             /* Sign of each element will be negative for non-real atoms.
 330              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 331              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 332              */
 333             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 334
 335             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 336             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 337             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 338
 339             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 340             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 341             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 342             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 343             j_coord_offsetA  = DIM*jnrA;
 344             j_coord_offsetB  = DIM*jnrB;
 345             j_coord_offsetC  = DIM*jnrC;
 346             j_coord_offsetD  = DIM*jnrD;
 347
 348             /* load j atom coordinates */
 349             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 350                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 351                                                  &jx0,&jy0,&jz0);
 352
 353             /* Calculate displacement vector */
 354             dx10             = _mm256_sub_pd(ix1,jx0);
 355             dy10             = _mm256_sub_pd(iy1,jy0);
 356             dz10             = _mm256_sub_pd(iz1,jz0);
 357             dx20             = _mm256_sub_pd(ix2,jx0);
 358             dy20             = _mm256_sub_pd(iy2,jy0);
 359             dz20             = _mm256_sub_pd(iz2,jz0);
 360             dx30             = _mm256_sub_pd(ix3,jx0);
 361             dy30             = _mm256_sub_pd(iy3,jy0);
 362             dz30             = _mm256_sub_pd(iz3,jz0);
 363
 364             /* Calculate squared distance and things based on it */
 365             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 366             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 367             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 368
 369             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 370             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 371             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 372
 373             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 374             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 375             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 376
 377             /* Load parameters for j particles */
 378             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 379                                                                  charge+jnrC+0,charge+jnrD+0);
 380
 381             fjx0             = _mm256_setzero_pd();
 382             fjy0             = _mm256_setzero_pd();
 383             fjz0             = _mm256_setzero_pd();
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq10             = _mm256_mul_pd(iq1,jq0);
 391
 392             /* COULOMB ELECTROSTATICS */
 393             velec            = _mm256_mul_pd(qq10,rinv10);
 394             felec            = _mm256_mul_pd(velec,rinvsq10);
 395
 396             /* Update potential sum for this i atom from the interaction with this j atom. */
 397             velec            = _mm256_andnot_pd(dummy_mask,velec);
 398             velecsum         = _mm256_add_pd(velecsum,velec);
 399
 400             fscal            = felec;
 401
 402             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 403
 404             /* Calculate temporary vectorial force */
 405             tx               = _mm256_mul_pd(fscal,dx10);
 406             ty               = _mm256_mul_pd(fscal,dy10);
 407             tz               = _mm256_mul_pd(fscal,dz10);
 408
 409             /* Update vectorial force */
 410             fix1             = _mm256_add_pd(fix1,tx);
 411             fiy1             = _mm256_add_pd(fiy1,ty);
 412             fiz1             = _mm256_add_pd(fiz1,tz);
 413
 414             fjx0             = _mm256_add_pd(fjx0,tx);
 415             fjy0             = _mm256_add_pd(fjy0,ty);
 416             fjz0             = _mm256_add_pd(fjz0,tz);
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             /* Compute parameters for interactions between i and j atoms */
 423             qq20             = _mm256_mul_pd(iq2,jq0);
 424
 425             /* COULOMB ELECTROSTATICS */
 426             velec            = _mm256_mul_pd(qq20,rinv20);
 427             felec            = _mm256_mul_pd(velec,rinvsq20);
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velec            = _mm256_andnot_pd(dummy_mask,velec);
 431             velecsum         = _mm256_add_pd(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 436
 437             /* Calculate temporary vectorial force */
 438             tx               = _mm256_mul_pd(fscal,dx20);
 439             ty               = _mm256_mul_pd(fscal,dy20);
 440             tz               = _mm256_mul_pd(fscal,dz20);
 441
 442             /* Update vectorial force */
 443             fix2             = _mm256_add_pd(fix2,tx);
 444             fiy2             = _mm256_add_pd(fiy2,ty);
 445             fiz2             = _mm256_add_pd(fiz2,tz);
 446
 447             fjx0             = _mm256_add_pd(fjx0,tx);
 448             fjy0             = _mm256_add_pd(fjy0,ty);
 449             fjz0             = _mm256_add_pd(fjz0,tz);
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             /* Compute parameters for interactions between i and j atoms */
 456             qq30             = _mm256_mul_pd(iq3,jq0);
 457
 458             /* COULOMB ELECTROSTATICS */
 459             velec            = _mm256_mul_pd(qq30,rinv30);
 460             felec            = _mm256_mul_pd(velec,rinvsq30);
 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,dx30);
 472             ty               = _mm256_mul_pd(fscal,dy30);
 473             tz               = _mm256_mul_pd(fscal,dz30);
 474
 475             /* Update vectorial force */
 476             fix3             = _mm256_add_pd(fix3,tx);
 477             fiy3             = _mm256_add_pd(fiy3,ty);
 478             fiz3             = _mm256_add_pd(fiz3,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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 485             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 486             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 487             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 488
 489             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 490
 491             /* Inner loop uses 84 flops */
 492         }
 493
 494         /* End of innermost loop */
 495
 496         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 497                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 498
 499         ggid                        = gid[iidx];
 500         /* Update potential energies */
 501         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 502
 503         /* Increment number of inner iterations */
 504         inneriter                  += j_index_end - j_index_start;
 505
 506         /* Outer loop uses 19 flops */
 507     }
 508
 509     /* Increment number of outer iterations */
 510     outeriter        += nri;
 511
 512     /* Update outer/inner flops */
 513
 514     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*84);
 515 }
 516 /*
 517  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_double
 518  * Electrostatics interaction: Coulomb
 519  * VdW interaction:            None
 520  * Geometry:                   Water4-Particle
 521  * Calculate force/pot:        Force
 522  */
 523 void
 524 nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_256_double
 525                     (t_nblist                    * gmx_restrict       nlist,
 526                      rvec                        * gmx_restrict          xx,
 527                      rvec                        * gmx_restrict          ff,
 528                      t_forcerec                  * gmx_restrict          fr,
 529                      t_mdatoms                   * gmx_restrict     mdatoms,
 530                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 531                      t_nrnb                      * gmx_restrict        nrnb)
 532 {
 533     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 534      * just 0 for non-waters.
 535      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 536      * jnr indices corresponding to data put in the four positions in the SIMD register.
 537      */
 538     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 539     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 540     int              jnrA,jnrB,jnrC,jnrD;
 541     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 542     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 543     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 544     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 545     real             rcutoff_scalar;
 546     real             *shiftvec,*fshift,*x,*f;
 547     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 548     real             scratch[4*DIM];
 549     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 550     real *           vdwioffsetptr1;
 551     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 552     real *           vdwioffsetptr2;
 553     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 554     real *           vdwioffsetptr3;
 555     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 556     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 557     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 558     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 559     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 560     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 561     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 562     real             *charge;
 563     __m256d          dummy_mask,cutoff_mask;
 564     __m128           tmpmask0,tmpmask1;
 565     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 566     __m256d          one     = _mm256_set1_pd(1.0);
 567     __m256d          two     = _mm256_set1_pd(2.0);
 568     x                = xx[0];
 569     f                = ff[0];
 570
 571     nri              = nlist->nri;
 572     iinr             = nlist->iinr;
 573     jindex           = nlist->jindex;
 574     jjnr             = nlist->jjnr;
 575     shiftidx         = nlist->shift;
 576     gid              = nlist->gid;
 577     shiftvec         = fr->shift_vec[0];
 578     fshift           = fr->fshift[0];
 579     facel            = _mm256_set1_pd(fr->epsfac);
 580     charge           = mdatoms->chargeA;
 581
 582     /* Setup water-specific parameters */
 583     inr              = nlist->iinr[0];
 584     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 585     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 586     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 587
 588     /* Avoid stupid compiler warnings */
 589     jnrA = jnrB = jnrC = jnrD = 0;
 590     j_coord_offsetA = 0;
 591     j_coord_offsetB = 0;
 592     j_coord_offsetC = 0;
 593     j_coord_offsetD = 0;
 594
 595     outeriter        = 0;
 596     inneriter        = 0;
 597
 598     for(iidx=0;iidx<4*DIM;iidx++)
 599     {
 600         scratch[iidx] = 0.0;
 601     }
 602
 603     /* Start outer loop over neighborlists */
 604     for(iidx=0; iidx<nri; iidx++)
 605     {
 606         /* Load shift vector for this list */
 607         i_shift_offset   = DIM*shiftidx[iidx];
 608
 609         /* Load limits for loop over neighbors */
 610         j_index_start    = jindex[iidx];
 611         j_index_end      = jindex[iidx+1];
 612
 613         /* Get outer coordinate index */
 614         inr              = iinr[iidx];
 615         i_coord_offset   = DIM*inr;
 616
 617         /* Load i particle coords and add shift vector */
 618         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 619                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 620
 621         fix1             = _mm256_setzero_pd();
 622         fiy1             = _mm256_setzero_pd();
 623         fiz1             = _mm256_setzero_pd();
 624         fix2             = _mm256_setzero_pd();
 625         fiy2             = _mm256_setzero_pd();
 626         fiz2             = _mm256_setzero_pd();
 627         fix3             = _mm256_setzero_pd();
 628         fiy3             = _mm256_setzero_pd();
 629         fiz3             = _mm256_setzero_pd();
 630
 631         /* Start inner kernel loop */
 632         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 633         {
 634
 635             /* Get j neighbor index, and coordinate index */
 636             jnrA             = jjnr[jidx];
 637             jnrB             = jjnr[jidx+1];
 638             jnrC             = jjnr[jidx+2];
 639             jnrD             = jjnr[jidx+3];
 640             j_coord_offsetA  = DIM*jnrA;
 641             j_coord_offsetB  = DIM*jnrB;
 642             j_coord_offsetC  = DIM*jnrC;
 643             j_coord_offsetD  = DIM*jnrD;
 644
 645             /* load j atom coordinates */
 646             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 647                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 648                                                  &jx0,&jy0,&jz0);
 649
 650             /* Calculate displacement vector */
 651             dx10             = _mm256_sub_pd(ix1,jx0);
 652             dy10             = _mm256_sub_pd(iy1,jy0);
 653             dz10             = _mm256_sub_pd(iz1,jz0);
 654             dx20             = _mm256_sub_pd(ix2,jx0);
 655             dy20             = _mm256_sub_pd(iy2,jy0);
 656             dz20             = _mm256_sub_pd(iz2,jz0);
 657             dx30             = _mm256_sub_pd(ix3,jx0);
 658             dy30             = _mm256_sub_pd(iy3,jy0);
 659             dz30             = _mm256_sub_pd(iz3,jz0);
 660
 661             /* Calculate squared distance and things based on it */
 662             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 663             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 664             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 665
 666             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 667             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 668             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 669
 670             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 671             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 672             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 673
 674             /* Load parameters for j particles */
 675             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 676                                                                  charge+jnrC+0,charge+jnrD+0);
 677
 678             fjx0             = _mm256_setzero_pd();
 679             fjy0             = _mm256_setzero_pd();
 680             fjz0             = _mm256_setzero_pd();
 681
 682             /**************************
 683              * CALCULATE INTERACTIONS *
 684              **************************/
 685
 686             /* Compute parameters for interactions between i and j atoms */
 687             qq10             = _mm256_mul_pd(iq1,jq0);
 688
 689             /* COULOMB ELECTROSTATICS */
 690             velec            = _mm256_mul_pd(qq10,rinv10);
 691             felec            = _mm256_mul_pd(velec,rinvsq10);
 692
 693             fscal            = felec;
 694
 695             /* Calculate temporary vectorial force */
 696             tx               = _mm256_mul_pd(fscal,dx10);
 697             ty               = _mm256_mul_pd(fscal,dy10);
 698             tz               = _mm256_mul_pd(fscal,dz10);
 699
 700             /* Update vectorial force */
 701             fix1             = _mm256_add_pd(fix1,tx);
 702             fiy1             = _mm256_add_pd(fiy1,ty);
 703             fiz1             = _mm256_add_pd(fiz1,tz);
 704
 705             fjx0             = _mm256_add_pd(fjx0,tx);
 706             fjy0             = _mm256_add_pd(fjy0,ty);
 707             fjz0             = _mm256_add_pd(fjz0,tz);
 708
 709             /**************************
 710              * CALCULATE INTERACTIONS *
 711              **************************/
 712
 713             /* Compute parameters for interactions between i and j atoms */
 714             qq20             = _mm256_mul_pd(iq2,jq0);
 715
 716             /* COULOMB ELECTROSTATICS */
 717             velec            = _mm256_mul_pd(qq20,rinv20);
 718             felec            = _mm256_mul_pd(velec,rinvsq20);
 719
 720             fscal            = felec;
 721
 722             /* Calculate temporary vectorial force */
 723             tx               = _mm256_mul_pd(fscal,dx20);
 724             ty               = _mm256_mul_pd(fscal,dy20);
 725             tz               = _mm256_mul_pd(fscal,dz20);
 726
 727             /* Update vectorial force */
 728             fix2             = _mm256_add_pd(fix2,tx);
 729             fiy2             = _mm256_add_pd(fiy2,ty);
 730             fiz2             = _mm256_add_pd(fiz2,tz);
 731
 732             fjx0             = _mm256_add_pd(fjx0,tx);
 733             fjy0             = _mm256_add_pd(fjy0,ty);
 734             fjz0             = _mm256_add_pd(fjz0,tz);
 735
 736             /**************************
 737              * CALCULATE INTERACTIONS *
 738              **************************/
 739
 740             /* Compute parameters for interactions between i and j atoms */
 741             qq30             = _mm256_mul_pd(iq3,jq0);
 742
 743             /* COULOMB ELECTROSTATICS */
 744             velec            = _mm256_mul_pd(qq30,rinv30);
 745             felec            = _mm256_mul_pd(velec,rinvsq30);
 746
 747             fscal            = felec;
 748
 749             /* Calculate temporary vectorial force */
 750             tx               = _mm256_mul_pd(fscal,dx30);
 751             ty               = _mm256_mul_pd(fscal,dy30);
 752             tz               = _mm256_mul_pd(fscal,dz30);
 753
 754             /* Update vectorial force */
 755             fix3             = _mm256_add_pd(fix3,tx);
 756             fiy3             = _mm256_add_pd(fiy3,ty);
 757             fiz3             = _mm256_add_pd(fiz3,tz);
 758
 759             fjx0             = _mm256_add_pd(fjx0,tx);
 760             fjy0             = _mm256_add_pd(fjy0,ty);
 761             fjz0             = _mm256_add_pd(fjz0,tz);
 762
 763             fjptrA             = f+j_coord_offsetA;
 764             fjptrB             = f+j_coord_offsetB;
 765             fjptrC             = f+j_coord_offsetC;
 766             fjptrD             = f+j_coord_offsetD;
 767
 768             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 769
 770             /* Inner loop uses 81 flops */
 771         }
 772
 773         if(jidx<j_index_end)
 774         {
 775
 776             /* Get j neighbor index, and coordinate index */
 777             jnrlistA         = jjnr[jidx];
 778             jnrlistB         = jjnr[jidx+1];
 779             jnrlistC         = jjnr[jidx+2];
 780             jnrlistD         = jjnr[jidx+3];
 781             /* Sign of each element will be negative for non-real atoms.
 782              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 783              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 784              */
 785             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 786
 787             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 788             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 789             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 790
 791             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 792             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 793             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 794             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 795             j_coord_offsetA  = DIM*jnrA;
 796             j_coord_offsetB  = DIM*jnrB;
 797             j_coord_offsetC  = DIM*jnrC;
 798             j_coord_offsetD  = DIM*jnrD;
 799
 800             /* load j atom coordinates */
 801             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 802                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 803                                                  &jx0,&jy0,&jz0);
 804
 805             /* Calculate displacement vector */
 806             dx10             = _mm256_sub_pd(ix1,jx0);
 807             dy10             = _mm256_sub_pd(iy1,jy0);
 808             dz10             = _mm256_sub_pd(iz1,jz0);
 809             dx20             = _mm256_sub_pd(ix2,jx0);
 810             dy20             = _mm256_sub_pd(iy2,jy0);
 811             dz20             = _mm256_sub_pd(iz2,jz0);
 812             dx30             = _mm256_sub_pd(ix3,jx0);
 813             dy30             = _mm256_sub_pd(iy3,jy0);
 814             dz30             = _mm256_sub_pd(iz3,jz0);
 815
 816             /* Calculate squared distance and things based on it */
 817             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 818             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 819             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 820
 821             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 822             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 823             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 824
 825             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 826             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 827             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 828
 829             /* Load parameters for j particles */
 830             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 831                                                                  charge+jnrC+0,charge+jnrD+0);
 832
 833             fjx0             = _mm256_setzero_pd();
 834             fjy0             = _mm256_setzero_pd();
 835             fjz0             = _mm256_setzero_pd();
 836
 837             /**************************
 838              * CALCULATE INTERACTIONS *
 839              **************************/
 840
 841             /* Compute parameters for interactions between i and j atoms */
 842             qq10             = _mm256_mul_pd(iq1,jq0);
 843
 844             /* COULOMB ELECTROSTATICS */
 845             velec            = _mm256_mul_pd(qq10,rinv10);
 846             felec            = _mm256_mul_pd(velec,rinvsq10);
 847
 848             fscal            = felec;
 849
 850             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm256_mul_pd(fscal,dx10);
 854             ty               = _mm256_mul_pd(fscal,dy10);
 855             tz               = _mm256_mul_pd(fscal,dz10);
 856
 857             /* Update vectorial force */
 858             fix1             = _mm256_add_pd(fix1,tx);
 859             fiy1             = _mm256_add_pd(fiy1,ty);
 860             fiz1             = _mm256_add_pd(fiz1,tz);
 861
 862             fjx0             = _mm256_add_pd(fjx0,tx);
 863             fjy0             = _mm256_add_pd(fjy0,ty);
 864             fjz0             = _mm256_add_pd(fjz0,tz);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq20             = _mm256_mul_pd(iq2,jq0);
 872
 873             /* COULOMB ELECTROSTATICS */
 874             velec            = _mm256_mul_pd(qq20,rinv20);
 875             felec            = _mm256_mul_pd(velec,rinvsq20);
 876
 877             fscal            = felec;
 878
 879             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 880
 881             /* Calculate temporary vectorial force */
 882             tx               = _mm256_mul_pd(fscal,dx20);
 883             ty               = _mm256_mul_pd(fscal,dy20);
 884             tz               = _mm256_mul_pd(fscal,dz20);
 885
 886             /* Update vectorial force */
 887             fix2             = _mm256_add_pd(fix2,tx);
 888             fiy2             = _mm256_add_pd(fiy2,ty);
 889             fiz2             = _mm256_add_pd(fiz2,tz);
 890
 891             fjx0             = _mm256_add_pd(fjx0,tx);
 892             fjy0             = _mm256_add_pd(fjy0,ty);
 893             fjz0             = _mm256_add_pd(fjz0,tz);
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             /* Compute parameters for interactions between i and j atoms */
 900             qq30             = _mm256_mul_pd(iq3,jq0);
 901
 902             /* COULOMB ELECTROSTATICS */
 903             velec            = _mm256_mul_pd(qq30,rinv30);
 904             felec            = _mm256_mul_pd(velec,rinvsq30);
 905
 906             fscal            = felec;
 907
 908             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 909
 910             /* Calculate temporary vectorial force */
 911             tx               = _mm256_mul_pd(fscal,dx30);
 912             ty               = _mm256_mul_pd(fscal,dy30);
 913             tz               = _mm256_mul_pd(fscal,dz30);
 914
 915             /* Update vectorial force */
 916             fix3             = _mm256_add_pd(fix3,tx);
 917             fiy3             = _mm256_add_pd(fiy3,ty);
 918             fiz3             = _mm256_add_pd(fiz3,tz);
 919
 920             fjx0             = _mm256_add_pd(fjx0,tx);
 921             fjy0             = _mm256_add_pd(fjy0,ty);
 922             fjz0             = _mm256_add_pd(fjz0,tz);
 923
 924             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 925             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 926             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 927             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 928
 929             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 930
 931             /* Inner loop uses 81 flops */
 932         }
 933
 934         /* End of innermost loop */
 935
 936         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 937                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 938
 939         /* Increment number of inner iterations */
 940         inneriter                  += j_index_end - j_index_start;
 941
 942         /* Outer loop uses 18 flops */
 943     }
 944
 945     /* Increment number of outer iterations */
 946     outeriter        += nri;
 947
 948     /* Update outer/inner flops */
 949
 950     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*81);
 951 }