src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 sse4_1_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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128d          dummy_mask,cutoff_mask;
  99     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 100     __m128d          one     = _mm_set1_pd(1.0);
 101     __m128d          two     = _mm_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_pd(fr->ic->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 122     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 123     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 150                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 151
 152         fix0             = _mm_setzero_pd();
 153         fiy0             = _mm_setzero_pd();
 154         fiz0             = _mm_setzero_pd();
 155         fix1             = _mm_setzero_pd();
 156         fiy1             = _mm_setzero_pd();
 157         fiz1             = _mm_setzero_pd();
 158         fix2             = _mm_setzero_pd();
 159         fiy2             = _mm_setzero_pd();
 160         fiz2             = _mm_setzero_pd();
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164         vvdwsum          = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184             dx10             = _mm_sub_pd(ix1,jx0);
 185             dy10             = _mm_sub_pd(iy1,jy0);
 186             dz10             = _mm_sub_pd(iz1,jz0);
 187             dx20             = _mm_sub_pd(ix2,jx0);
 188             dy20             = _mm_sub_pd(iy2,jy0);
 189             dz20             = _mm_sub_pd(iz2,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 193             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 194             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 195
 196             rinv00           = sse41_invsqrt_d(rsq00);
 197             rinv10           = sse41_invsqrt_d(rsq10);
 198             rinv20           = sse41_invsqrt_d(rsq20);
 199
 200             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 201             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 202             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 203
 204             /* Load parameters for j particles */
 205             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 206             vdwjidx0A        = 2*vdwtype[jnrA+0];
 207             vdwjidx0B        = 2*vdwtype[jnrB+0];
 208
 209             fjx0             = _mm_setzero_pd();
 210             fjy0             = _mm_setzero_pd();
 211             fjz0             = _mm_setzero_pd();
 212
 213             /**************************
 214              * CALCULATE INTERACTIONS *
 215              **************************/
 216
 217             /* Compute parameters for interactions between i and j atoms */
 218             qq00             = _mm_mul_pd(iq0,jq0);
 219             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 220                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 221
 222             /* COULOMB ELECTROSTATICS */
 223             velec            = _mm_mul_pd(qq00,rinv00);
 224             felec            = _mm_mul_pd(velec,rinvsq00);
 225
 226             /* LENNARD-JONES DISPERSION/REPULSION */
 227
 228             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 229             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 230             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 231             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 232             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 233
 234             /* Update potential sum for this i atom from the interaction with this j atom. */
 235             velecsum         = _mm_add_pd(velecsum,velec);
 236             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 237
 238             fscal            = _mm_add_pd(felec,fvdw);
 239
 240             /* Calculate temporary vectorial force */
 241             tx               = _mm_mul_pd(fscal,dx00);
 242             ty               = _mm_mul_pd(fscal,dy00);
 243             tz               = _mm_mul_pd(fscal,dz00);
 244
 245             /* Update vectorial force */
 246             fix0             = _mm_add_pd(fix0,tx);
 247             fiy0             = _mm_add_pd(fiy0,ty);
 248             fiz0             = _mm_add_pd(fiz0,tz);
 249
 250             fjx0             = _mm_add_pd(fjx0,tx);
 251             fjy0             = _mm_add_pd(fjy0,ty);
 252             fjz0             = _mm_add_pd(fjz0,tz);
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             qq10             = _mm_mul_pd(iq1,jq0);
 260
 261             /* COULOMB ELECTROSTATICS */
 262             velec            = _mm_mul_pd(qq10,rinv10);
 263             felec            = _mm_mul_pd(velec,rinvsq10);
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             velecsum         = _mm_add_pd(velecsum,velec);
 267
 268             fscal            = felec;
 269
 270             /* Calculate temporary vectorial force */
 271             tx               = _mm_mul_pd(fscal,dx10);
 272             ty               = _mm_mul_pd(fscal,dy10);
 273             tz               = _mm_mul_pd(fscal,dz10);
 274
 275             /* Update vectorial force */
 276             fix1             = _mm_add_pd(fix1,tx);
 277             fiy1             = _mm_add_pd(fiy1,ty);
 278             fiz1             = _mm_add_pd(fiz1,tz);
 279
 280             fjx0             = _mm_add_pd(fjx0,tx);
 281             fjy0             = _mm_add_pd(fjy0,ty);
 282             fjz0             = _mm_add_pd(fjz0,tz);
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             qq20             = _mm_mul_pd(iq2,jq0);
 290
 291             /* COULOMB ELECTROSTATICS */
 292             velec            = _mm_mul_pd(qq20,rinv20);
 293             felec            = _mm_mul_pd(velec,rinvsq20);
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velecsum         = _mm_add_pd(velecsum,velec);
 297
 298             fscal            = felec;
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm_mul_pd(fscal,dx20);
 302             ty               = _mm_mul_pd(fscal,dy20);
 303             tz               = _mm_mul_pd(fscal,dz20);
 304
 305             /* Update vectorial force */
 306             fix2             = _mm_add_pd(fix2,tx);
 307             fiy2             = _mm_add_pd(fiy2,ty);
 308             fiz2             = _mm_add_pd(fiz2,tz);
 309
 310             fjx0             = _mm_add_pd(fjx0,tx);
 311             fjy0             = _mm_add_pd(fjy0,ty);
 312             fjz0             = _mm_add_pd(fjz0,tz);
 313
 314             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 315
 316             /* Inner loop uses 99 flops */
 317         }
 318
 319         if(jidx<j_index_end)
 320         {
 321
 322             jnrA             = jjnr[jidx];
 323             j_coord_offsetA  = DIM*jnrA;
 324
 325             /* load j atom coordinates */
 326             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 327                                               &jx0,&jy0,&jz0);
 328
 329             /* Calculate displacement vector */
 330             dx00             = _mm_sub_pd(ix0,jx0);
 331             dy00             = _mm_sub_pd(iy0,jy0);
 332             dz00             = _mm_sub_pd(iz0,jz0);
 333             dx10             = _mm_sub_pd(ix1,jx0);
 334             dy10             = _mm_sub_pd(iy1,jy0);
 335             dz10             = _mm_sub_pd(iz1,jz0);
 336             dx20             = _mm_sub_pd(ix2,jx0);
 337             dy20             = _mm_sub_pd(iy2,jy0);
 338             dz20             = _mm_sub_pd(iz2,jz0);
 339
 340             /* Calculate squared distance and things based on it */
 341             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 342             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 343             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 344
 345             rinv00           = sse41_invsqrt_d(rsq00);
 346             rinv10           = sse41_invsqrt_d(rsq10);
 347             rinv20           = sse41_invsqrt_d(rsq20);
 348
 349             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 350             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 351             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 352
 353             /* Load parameters for j particles */
 354             jq0              = _mm_load_sd(charge+jnrA+0);
 355             vdwjidx0A        = 2*vdwtype[jnrA+0];
 356
 357             fjx0             = _mm_setzero_pd();
 358             fjy0             = _mm_setzero_pd();
 359             fjz0             = _mm_setzero_pd();
 360
 361             /**************************
 362              * CALCULATE INTERACTIONS *
 363              **************************/
 364
 365             /* Compute parameters for interactions between i and j atoms */
 366             qq00             = _mm_mul_pd(iq0,jq0);
 367             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 368
 369             /* COULOMB ELECTROSTATICS */
 370             velec            = _mm_mul_pd(qq00,rinv00);
 371             felec            = _mm_mul_pd(velec,rinvsq00);
 372
 373             /* LENNARD-JONES DISPERSION/REPULSION */
 374
 375             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 376             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 377             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 378             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 379             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 380
 381             /* Update potential sum for this i atom from the interaction with this j atom. */
 382             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 383             velecsum         = _mm_add_pd(velecsum,velec);
 384             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 385             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 386
 387             fscal            = _mm_add_pd(felec,fvdw);
 388
 389             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 390
 391             /* Calculate temporary vectorial force */
 392             tx               = _mm_mul_pd(fscal,dx00);
 393             ty               = _mm_mul_pd(fscal,dy00);
 394             tz               = _mm_mul_pd(fscal,dz00);
 395
 396             /* Update vectorial force */
 397             fix0             = _mm_add_pd(fix0,tx);
 398             fiy0             = _mm_add_pd(fiy0,ty);
 399             fiz0             = _mm_add_pd(fiz0,tz);
 400
 401             fjx0             = _mm_add_pd(fjx0,tx);
 402             fjy0             = _mm_add_pd(fjy0,ty);
 403             fjz0             = _mm_add_pd(fjz0,tz);
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             /* Compute parameters for interactions between i and j atoms */
 410             qq10             = _mm_mul_pd(iq1,jq0);
 411
 412             /* COULOMB ELECTROSTATICS */
 413             velec            = _mm_mul_pd(qq10,rinv10);
 414             felec            = _mm_mul_pd(velec,rinvsq10);
 415
 416             /* Update potential sum for this i atom from the interaction with this j atom. */
 417             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 418             velecsum         = _mm_add_pd(velecsum,velec);
 419
 420             fscal            = felec;
 421
 422             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 423
 424             /* Calculate temporary vectorial force */
 425             tx               = _mm_mul_pd(fscal,dx10);
 426             ty               = _mm_mul_pd(fscal,dy10);
 427             tz               = _mm_mul_pd(fscal,dz10);
 428
 429             /* Update vectorial force */
 430             fix1             = _mm_add_pd(fix1,tx);
 431             fiy1             = _mm_add_pd(fiy1,ty);
 432             fiz1             = _mm_add_pd(fiz1,tz);
 433
 434             fjx0             = _mm_add_pd(fjx0,tx);
 435             fjy0             = _mm_add_pd(fjy0,ty);
 436             fjz0             = _mm_add_pd(fjz0,tz);
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             /* Compute parameters for interactions between i and j atoms */
 443             qq20             = _mm_mul_pd(iq2,jq0);
 444
 445             /* COULOMB ELECTROSTATICS */
 446             velec            = _mm_mul_pd(qq20,rinv20);
 447             felec            = _mm_mul_pd(velec,rinvsq20);
 448
 449             /* Update potential sum for this i atom from the interaction with this j atom. */
 450             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 451             velecsum         = _mm_add_pd(velecsum,velec);
 452
 453             fscal            = felec;
 454
 455             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 456
 457             /* Calculate temporary vectorial force */
 458             tx               = _mm_mul_pd(fscal,dx20);
 459             ty               = _mm_mul_pd(fscal,dy20);
 460             tz               = _mm_mul_pd(fscal,dz20);
 461
 462             /* Update vectorial force */
 463             fix2             = _mm_add_pd(fix2,tx);
 464             fiy2             = _mm_add_pd(fiy2,ty);
 465             fiz2             = _mm_add_pd(fiz2,tz);
 466
 467             fjx0             = _mm_add_pd(fjx0,tx);
 468             fjy0             = _mm_add_pd(fjy0,ty);
 469             fjz0             = _mm_add_pd(fjz0,tz);
 470
 471             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 472
 473             /* Inner loop uses 99 flops */
 474         }
 475
 476         /* End of innermost loop */
 477
 478         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 479                                               f+i_coord_offset,fshift+i_shift_offset);
 480
 481         ggid                        = gid[iidx];
 482         /* Update potential energies */
 483         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 484         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 485
 486         /* Increment number of inner iterations */
 487         inneriter                  += j_index_end - j_index_start;
 488
 489         /* Outer loop uses 20 flops */
 490     }
 491
 492     /* Increment number of outer iterations */
 493     outeriter        += nri;
 494
 495     /* Update outer/inner flops */
 496
 497     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*99);
 498 }
 499 /*
 500  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_double
 501  * Electrostatics interaction: Coulomb
 502  * VdW interaction:            LennardJones
 503  * Geometry:                   Water3-Particle
 504  * Calculate force/pot:        Force
 505  */
 506 void
 507 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_double
 508                     (t_nblist                    * gmx_restrict       nlist,
 509                      rvec                        * gmx_restrict          xx,
 510                      rvec                        * gmx_restrict          ff,
 511                      struct t_forcerec           * gmx_restrict          fr,
 512                      t_mdatoms                   * gmx_restrict     mdatoms,
 513                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 514                      t_nrnb                      * gmx_restrict        nrnb)
 515 {
 516     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 517      * just 0 for non-waters.
 518      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 519      * jnr indices corresponding to data put in the four positions in the SIMD register.
 520      */
 521     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 522     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 523     int              jnrA,jnrB;
 524     int              j_coord_offsetA,j_coord_offsetB;
 525     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 526     real             rcutoff_scalar;
 527     real             *shiftvec,*fshift,*x,*f;
 528     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 529     int              vdwioffset0;
 530     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 531     int              vdwioffset1;
 532     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 533     int              vdwioffset2;
 534     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 535     int              vdwjidx0A,vdwjidx0B;
 536     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 537     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 538     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 539     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 540     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 541     real             *charge;
 542     int              nvdwtype;
 543     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 544     int              *vdwtype;
 545     real             *vdwparam;
 546     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 547     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 548     __m128d          dummy_mask,cutoff_mask;
 549     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 550     __m128d          one     = _mm_set1_pd(1.0);
 551     __m128d          two     = _mm_set1_pd(2.0);
 552     x                = xx[0];
 553     f                = ff[0];
 554
 555     nri              = nlist->nri;
 556     iinr             = nlist->iinr;
 557     jindex           = nlist->jindex;
 558     jjnr             = nlist->jjnr;
 559     shiftidx         = nlist->shift;
 560     gid              = nlist->gid;
 561     shiftvec         = fr->shift_vec[0];
 562     fshift           = fr->fshift[0];
 563     facel            = _mm_set1_pd(fr->ic->epsfac);
 564     charge           = mdatoms->chargeA;
 565     nvdwtype         = fr->ntype;
 566     vdwparam         = fr->nbfp;
 567     vdwtype          = mdatoms->typeA;
 568
 569     /* Setup water-specific parameters */
 570     inr              = nlist->iinr[0];
 571     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 572     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 573     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 574     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 575
 576     /* Avoid stupid compiler warnings */
 577     jnrA = jnrB = 0;
 578     j_coord_offsetA = 0;
 579     j_coord_offsetB = 0;
 580
 581     outeriter        = 0;
 582     inneriter        = 0;
 583
 584     /* Start outer loop over neighborlists */
 585     for(iidx=0; iidx<nri; iidx++)
 586     {
 587         /* Load shift vector for this list */
 588         i_shift_offset   = DIM*shiftidx[iidx];
 589
 590         /* Load limits for loop over neighbors */
 591         j_index_start    = jindex[iidx];
 592         j_index_end      = jindex[iidx+1];
 593
 594         /* Get outer coordinate index */
 595         inr              = iinr[iidx];
 596         i_coord_offset   = DIM*inr;
 597
 598         /* Load i particle coords and add shift vector */
 599         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 600                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 601
 602         fix0             = _mm_setzero_pd();
 603         fiy0             = _mm_setzero_pd();
 604         fiz0             = _mm_setzero_pd();
 605         fix1             = _mm_setzero_pd();
 606         fiy1             = _mm_setzero_pd();
 607         fiz1             = _mm_setzero_pd();
 608         fix2             = _mm_setzero_pd();
 609         fiy2             = _mm_setzero_pd();
 610         fiz2             = _mm_setzero_pd();
 611
 612         /* Start inner kernel loop */
 613         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 614         {
 615
 616             /* Get j neighbor index, and coordinate index */
 617             jnrA             = jjnr[jidx];
 618             jnrB             = jjnr[jidx+1];
 619             j_coord_offsetA  = DIM*jnrA;
 620             j_coord_offsetB  = DIM*jnrB;
 621
 622             /* load j atom coordinates */
 623             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 624                                               &jx0,&jy0,&jz0);
 625
 626             /* Calculate displacement vector */
 627             dx00             = _mm_sub_pd(ix0,jx0);
 628             dy00             = _mm_sub_pd(iy0,jy0);
 629             dz00             = _mm_sub_pd(iz0,jz0);
 630             dx10             = _mm_sub_pd(ix1,jx0);
 631             dy10             = _mm_sub_pd(iy1,jy0);
 632             dz10             = _mm_sub_pd(iz1,jz0);
 633             dx20             = _mm_sub_pd(ix2,jx0);
 634             dy20             = _mm_sub_pd(iy2,jy0);
 635             dz20             = _mm_sub_pd(iz2,jz0);
 636
 637             /* Calculate squared distance and things based on it */
 638             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 639             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 640             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 641
 642             rinv00           = sse41_invsqrt_d(rsq00);
 643             rinv10           = sse41_invsqrt_d(rsq10);
 644             rinv20           = sse41_invsqrt_d(rsq20);
 645
 646             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 647             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 648             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 649
 650             /* Load parameters for j particles */
 651             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 652             vdwjidx0A        = 2*vdwtype[jnrA+0];
 653             vdwjidx0B        = 2*vdwtype[jnrB+0];
 654
 655             fjx0             = _mm_setzero_pd();
 656             fjy0             = _mm_setzero_pd();
 657             fjz0             = _mm_setzero_pd();
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             /* Compute parameters for interactions between i and j atoms */
 664             qq00             = _mm_mul_pd(iq0,jq0);
 665             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 666                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 667
 668             /* COULOMB ELECTROSTATICS */
 669             velec            = _mm_mul_pd(qq00,rinv00);
 670             felec            = _mm_mul_pd(velec,rinvsq00);
 671
 672             /* LENNARD-JONES DISPERSION/REPULSION */
 673
 674             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 675             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 676
 677             fscal            = _mm_add_pd(felec,fvdw);
 678
 679             /* Calculate temporary vectorial force */
 680             tx               = _mm_mul_pd(fscal,dx00);
 681             ty               = _mm_mul_pd(fscal,dy00);
 682             tz               = _mm_mul_pd(fscal,dz00);
 683
 684             /* Update vectorial force */
 685             fix0             = _mm_add_pd(fix0,tx);
 686             fiy0             = _mm_add_pd(fiy0,ty);
 687             fiz0             = _mm_add_pd(fiz0,tz);
 688
 689             fjx0             = _mm_add_pd(fjx0,tx);
 690             fjy0             = _mm_add_pd(fjy0,ty);
 691             fjz0             = _mm_add_pd(fjz0,tz);
 692
 693             /**************************
 694              * CALCULATE INTERACTIONS *
 695              **************************/
 696
 697             /* Compute parameters for interactions between i and j atoms */
 698             qq10             = _mm_mul_pd(iq1,jq0);
 699
 700             /* COULOMB ELECTROSTATICS */
 701             velec            = _mm_mul_pd(qq10,rinv10);
 702             felec            = _mm_mul_pd(velec,rinvsq10);
 703
 704             fscal            = felec;
 705
 706             /* Calculate temporary vectorial force */
 707             tx               = _mm_mul_pd(fscal,dx10);
 708             ty               = _mm_mul_pd(fscal,dy10);
 709             tz               = _mm_mul_pd(fscal,dz10);
 710
 711             /* Update vectorial force */
 712             fix1             = _mm_add_pd(fix1,tx);
 713             fiy1             = _mm_add_pd(fiy1,ty);
 714             fiz1             = _mm_add_pd(fiz1,tz);
 715
 716             fjx0             = _mm_add_pd(fjx0,tx);
 717             fjy0             = _mm_add_pd(fjy0,ty);
 718             fjz0             = _mm_add_pd(fjz0,tz);
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             /* Compute parameters for interactions between i and j atoms */
 725             qq20             = _mm_mul_pd(iq2,jq0);
 726
 727             /* COULOMB ELECTROSTATICS */
 728             velec            = _mm_mul_pd(qq20,rinv20);
 729             felec            = _mm_mul_pd(velec,rinvsq20);
 730
 731             fscal            = felec;
 732
 733             /* Calculate temporary vectorial force */
 734             tx               = _mm_mul_pd(fscal,dx20);
 735             ty               = _mm_mul_pd(fscal,dy20);
 736             tz               = _mm_mul_pd(fscal,dz20);
 737
 738             /* Update vectorial force */
 739             fix2             = _mm_add_pd(fix2,tx);
 740             fiy2             = _mm_add_pd(fiy2,ty);
 741             fiz2             = _mm_add_pd(fiz2,tz);
 742
 743             fjx0             = _mm_add_pd(fjx0,tx);
 744             fjy0             = _mm_add_pd(fjy0,ty);
 745             fjz0             = _mm_add_pd(fjz0,tz);
 746
 747             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 748
 749             /* Inner loop uses 91 flops */
 750         }
 751
 752         if(jidx<j_index_end)
 753         {
 754
 755             jnrA             = jjnr[jidx];
 756             j_coord_offsetA  = DIM*jnrA;
 757
 758             /* load j atom coordinates */
 759             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 760                                               &jx0,&jy0,&jz0);
 761
 762             /* Calculate displacement vector */
 763             dx00             = _mm_sub_pd(ix0,jx0);
 764             dy00             = _mm_sub_pd(iy0,jy0);
 765             dz00             = _mm_sub_pd(iz0,jz0);
 766             dx10             = _mm_sub_pd(ix1,jx0);
 767             dy10             = _mm_sub_pd(iy1,jy0);
 768             dz10             = _mm_sub_pd(iz1,jz0);
 769             dx20             = _mm_sub_pd(ix2,jx0);
 770             dy20             = _mm_sub_pd(iy2,jy0);
 771             dz20             = _mm_sub_pd(iz2,jz0);
 772
 773             /* Calculate squared distance and things based on it */
 774             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 775             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 776             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 777
 778             rinv00           = sse41_invsqrt_d(rsq00);
 779             rinv10           = sse41_invsqrt_d(rsq10);
 780             rinv20           = sse41_invsqrt_d(rsq20);
 781
 782             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 783             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 784             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 785
 786             /* Load parameters for j particles */
 787             jq0              = _mm_load_sd(charge+jnrA+0);
 788             vdwjidx0A        = 2*vdwtype[jnrA+0];
 789
 790             fjx0             = _mm_setzero_pd();
 791             fjy0             = _mm_setzero_pd();
 792             fjz0             = _mm_setzero_pd();
 793
 794             /**************************
 795              * CALCULATE INTERACTIONS *
 796              **************************/
 797
 798             /* Compute parameters for interactions between i and j atoms */
 799             qq00             = _mm_mul_pd(iq0,jq0);
 800             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 801
 802             /* COULOMB ELECTROSTATICS */
 803             velec            = _mm_mul_pd(qq00,rinv00);
 804             felec            = _mm_mul_pd(velec,rinvsq00);
 805
 806             /* LENNARD-JONES DISPERSION/REPULSION */
 807
 808             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 809             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 810
 811             fscal            = _mm_add_pd(felec,fvdw);
 812
 813             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 814
 815             /* Calculate temporary vectorial force */
 816             tx               = _mm_mul_pd(fscal,dx00);
 817             ty               = _mm_mul_pd(fscal,dy00);
 818             tz               = _mm_mul_pd(fscal,dz00);
 819
 820             /* Update vectorial force */
 821             fix0             = _mm_add_pd(fix0,tx);
 822             fiy0             = _mm_add_pd(fiy0,ty);
 823             fiz0             = _mm_add_pd(fiz0,tz);
 824
 825             fjx0             = _mm_add_pd(fjx0,tx);
 826             fjy0             = _mm_add_pd(fjy0,ty);
 827             fjz0             = _mm_add_pd(fjz0,tz);
 828
 829             /**************************
 830              * CALCULATE INTERACTIONS *
 831              **************************/
 832
 833             /* Compute parameters for interactions between i and j atoms */
 834             qq10             = _mm_mul_pd(iq1,jq0);
 835
 836             /* COULOMB ELECTROSTATICS */
 837             velec            = _mm_mul_pd(qq10,rinv10);
 838             felec            = _mm_mul_pd(velec,rinvsq10);
 839
 840             fscal            = felec;
 841
 842             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 843
 844             /* Calculate temporary vectorial force */
 845             tx               = _mm_mul_pd(fscal,dx10);
 846             ty               = _mm_mul_pd(fscal,dy10);
 847             tz               = _mm_mul_pd(fscal,dz10);
 848
 849             /* Update vectorial force */
 850             fix1             = _mm_add_pd(fix1,tx);
 851             fiy1             = _mm_add_pd(fiy1,ty);
 852             fiz1             = _mm_add_pd(fiz1,tz);
 853
 854             fjx0             = _mm_add_pd(fjx0,tx);
 855             fjy0             = _mm_add_pd(fjy0,ty);
 856             fjz0             = _mm_add_pd(fjz0,tz);
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             /* Compute parameters for interactions between i and j atoms */
 863             qq20             = _mm_mul_pd(iq2,jq0);
 864
 865             /* COULOMB ELECTROSTATICS */
 866             velec            = _mm_mul_pd(qq20,rinv20);
 867             felec            = _mm_mul_pd(velec,rinvsq20);
 868
 869             fscal            = felec;
 870
 871             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 872
 873             /* Calculate temporary vectorial force */
 874             tx               = _mm_mul_pd(fscal,dx20);
 875             ty               = _mm_mul_pd(fscal,dy20);
 876             tz               = _mm_mul_pd(fscal,dz20);
 877
 878             /* Update vectorial force */
 879             fix2             = _mm_add_pd(fix2,tx);
 880             fiy2             = _mm_add_pd(fiy2,ty);
 881             fiz2             = _mm_add_pd(fiz2,tz);
 882
 883             fjx0             = _mm_add_pd(fjx0,tx);
 884             fjy0             = _mm_add_pd(fjy0,ty);
 885             fjz0             = _mm_add_pd(fjz0,tz);
 886
 887             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 888
 889             /* Inner loop uses 91 flops */
 890         }
 891
 892         /* End of innermost loop */
 893
 894         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 895                                               f+i_coord_offset,fshift+i_shift_offset);
 896
 897         /* Increment number of inner iterations */
 898         inneriter                  += j_index_end - j_index_start;
 899
 900         /* Outer loop uses 18 flops */
 901     }
 902
 903     /* Increment number of outer iterations */
 904     outeriter        += nri;
 905
 906     /* Update outer/inner flops */
 907
 908     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*91);
 909 }