src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwLJ_GeomW4P1_avx_128_fma_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 avx_128_fma_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_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_avx_128_fma_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              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128d          dummy_mask,cutoff_mask;
 102     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 103     __m128d          one     = _mm_set1_pd(1.0);
 104     __m128d          two     = _mm_set1_pd(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_pd(fr->ic->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* Setup water-specific parameters */
 123     inr              = nlist->iinr[0];
 124     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 125     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 126     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 127     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 153                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 154
 155         fix0             = _mm_setzero_pd();
 156         fiy0             = _mm_setzero_pd();
 157         fiz0             = _mm_setzero_pd();
 158         fix1             = _mm_setzero_pd();
 159         fiy1             = _mm_setzero_pd();
 160         fiz1             = _mm_setzero_pd();
 161         fix2             = _mm_setzero_pd();
 162         fiy2             = _mm_setzero_pd();
 163         fiz2             = _mm_setzero_pd();
 164         fix3             = _mm_setzero_pd();
 165         fiy3             = _mm_setzero_pd();
 166         fiz3             = _mm_setzero_pd();
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm_setzero_pd();
 170         vvdwsum          = _mm_setzero_pd();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             j_coord_offsetA  = DIM*jnrA;
 180             j_coord_offsetB  = DIM*jnrB;
 181
 182             /* load j atom coordinates */
 183             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 184                                               &jx0,&jy0,&jz0);
 185
 186             /* Calculate displacement vector */
 187             dx00             = _mm_sub_pd(ix0,jx0);
 188             dy00             = _mm_sub_pd(iy0,jy0);
 189             dz00             = _mm_sub_pd(iz0,jz0);
 190             dx10             = _mm_sub_pd(ix1,jx0);
 191             dy10             = _mm_sub_pd(iy1,jy0);
 192             dz10             = _mm_sub_pd(iz1,jz0);
 193             dx20             = _mm_sub_pd(ix2,jx0);
 194             dy20             = _mm_sub_pd(iy2,jy0);
 195             dz20             = _mm_sub_pd(iz2,jz0);
 196             dx30             = _mm_sub_pd(ix3,jx0);
 197             dy30             = _mm_sub_pd(iy3,jy0);
 198             dz30             = _mm_sub_pd(iz3,jz0);
 199
 200             /* Calculate squared distance and things based on it */
 201             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 202             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 203             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 204             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 205
 206             rinv10           = avx128fma_invsqrt_d(rsq10);
 207             rinv20           = avx128fma_invsqrt_d(rsq20);
 208             rinv30           = avx128fma_invsqrt_d(rsq30);
 209
 210             rinvsq00         = avx128fma_inv_d(rsq00);
 211             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 212             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 213             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 214
 215             /* Load parameters for j particles */
 216             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 217             vdwjidx0A        = 2*vdwtype[jnrA+0];
 218             vdwjidx0B        = 2*vdwtype[jnrB+0];
 219
 220             fjx0             = _mm_setzero_pd();
 221             fjy0             = _mm_setzero_pd();
 222             fjz0             = _mm_setzero_pd();
 223
 224             /**************************
 225              * CALCULATE INTERACTIONS *
 226              **************************/
 227
 228             /* Compute parameters for interactions between i and j atoms */
 229             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 230                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 231
 232             /* LENNARD-JONES DISPERSION/REPULSION */
 233
 234             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 235             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 236             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 237             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 238             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 239
 240             /* Update potential sum for this i atom from the interaction with this j atom. */
 241             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 242
 243             fscal            = fvdw;
 244
 245             /* Update vectorial force */
 246             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 247             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 248             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 249
 250             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 251             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 252             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 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             /* Update vectorial force */
 271             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 272             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 273             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 274
 275             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 276             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 277             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             /* Compute parameters for interactions between i and j atoms */
 284             qq20             = _mm_mul_pd(iq2,jq0);
 285
 286             /* COULOMB ELECTROSTATICS */
 287             velec            = _mm_mul_pd(qq20,rinv20);
 288             felec            = _mm_mul_pd(velec,rinvsq20);
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             velecsum         = _mm_add_pd(velecsum,velec);
 292
 293             fscal            = felec;
 294
 295             /* Update vectorial force */
 296             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 297             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 298             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 299
 300             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 301             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 302             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq30             = _mm_mul_pd(iq3,jq0);
 310
 311             /* COULOMB ELECTROSTATICS */
 312             velec            = _mm_mul_pd(qq30,rinv30);
 313             felec            = _mm_mul_pd(velec,rinvsq30);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velecsum         = _mm_add_pd(velecsum,velec);
 317
 318             fscal            = felec;
 319
 320             /* Update vectorial force */
 321             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 322             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 323             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 324
 325             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 326             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 327             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 328
 329             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 330
 331             /* Inner loop uses 131 flops */
 332         }
 333
 334         if(jidx<j_index_end)
 335         {
 336
 337             jnrA             = jjnr[jidx];
 338             j_coord_offsetA  = DIM*jnrA;
 339
 340             /* load j atom coordinates */
 341             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 342                                               &jx0,&jy0,&jz0);
 343
 344             /* Calculate displacement vector */
 345             dx00             = _mm_sub_pd(ix0,jx0);
 346             dy00             = _mm_sub_pd(iy0,jy0);
 347             dz00             = _mm_sub_pd(iz0,jz0);
 348             dx10             = _mm_sub_pd(ix1,jx0);
 349             dy10             = _mm_sub_pd(iy1,jy0);
 350             dz10             = _mm_sub_pd(iz1,jz0);
 351             dx20             = _mm_sub_pd(ix2,jx0);
 352             dy20             = _mm_sub_pd(iy2,jy0);
 353             dz20             = _mm_sub_pd(iz2,jz0);
 354             dx30             = _mm_sub_pd(ix3,jx0);
 355             dy30             = _mm_sub_pd(iy3,jy0);
 356             dz30             = _mm_sub_pd(iz3,jz0);
 357
 358             /* Calculate squared distance and things based on it */
 359             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 360             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 361             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 362             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 363
 364             rinv10           = avx128fma_invsqrt_d(rsq10);
 365             rinv20           = avx128fma_invsqrt_d(rsq20);
 366             rinv30           = avx128fma_invsqrt_d(rsq30);
 367
 368             rinvsq00         = avx128fma_inv_d(rsq00);
 369             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 370             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 371             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 372
 373             /* Load parameters for j particles */
 374             jq0              = _mm_load_sd(charge+jnrA+0);
 375             vdwjidx0A        = 2*vdwtype[jnrA+0];
 376
 377             fjx0             = _mm_setzero_pd();
 378             fjy0             = _mm_setzero_pd();
 379             fjz0             = _mm_setzero_pd();
 380
 381             /**************************
 382              * CALCULATE INTERACTIONS *
 383              **************************/
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 387
 388             /* LENNARD-JONES DISPERSION/REPULSION */
 389
 390             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 391             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 392             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 393             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 394             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 395
 396             /* Update potential sum for this i atom from the interaction with this j atom. */
 397             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 398             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 399
 400             fscal            = fvdw;
 401
 402             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 403
 404             /* Update vectorial force */
 405             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 406             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 407             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 408
 409             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 410             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 411             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 412
 413             /**************************
 414              * CALCULATE INTERACTIONS *
 415              **************************/
 416
 417             /* Compute parameters for interactions between i and j atoms */
 418             qq10             = _mm_mul_pd(iq1,jq0);
 419
 420             /* COULOMB ELECTROSTATICS */
 421             velec            = _mm_mul_pd(qq10,rinv10);
 422             felec            = _mm_mul_pd(velec,rinvsq10);
 423
 424             /* Update potential sum for this i atom from the interaction with this j atom. */
 425             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 426             velecsum         = _mm_add_pd(velecsum,velec);
 427
 428             fscal            = felec;
 429
 430             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 431
 432             /* Update vectorial force */
 433             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 434             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 435             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 436
 437             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 438             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 439             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 440
 441             /**************************
 442              * CALCULATE INTERACTIONS *
 443              **************************/
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq20             = _mm_mul_pd(iq2,jq0);
 447
 448             /* COULOMB ELECTROSTATICS */
 449             velec            = _mm_mul_pd(qq20,rinv20);
 450             felec            = _mm_mul_pd(velec,rinvsq20);
 451
 452             /* Update potential sum for this i atom from the interaction with this j atom. */
 453             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 454             velecsum         = _mm_add_pd(velecsum,velec);
 455
 456             fscal            = felec;
 457
 458             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 459
 460             /* Update vectorial force */
 461             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 462             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 463             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 464
 465             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 466             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 467             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 468
 469             /**************************
 470              * CALCULATE INTERACTIONS *
 471              **************************/
 472
 473             /* Compute parameters for interactions between i and j atoms */
 474             qq30             = _mm_mul_pd(iq3,jq0);
 475
 476             /* COULOMB ELECTROSTATICS */
 477             velec            = _mm_mul_pd(qq30,rinv30);
 478             felec            = _mm_mul_pd(velec,rinvsq30);
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 482             velecsum         = _mm_add_pd(velecsum,velec);
 483
 484             fscal            = felec;
 485
 486             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 487
 488             /* Update vectorial force */
 489             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 490             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 491             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 492
 493             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 494             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 495             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 496
 497             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 498
 499             /* Inner loop uses 131 flops */
 500         }
 501
 502         /* End of innermost loop */
 503
 504         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 505                                               f+i_coord_offset,fshift+i_shift_offset);
 506
 507         ggid                        = gid[iidx];
 508         /* Update potential energies */
 509         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 510         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 511
 512         /* Increment number of inner iterations */
 513         inneriter                  += j_index_end - j_index_start;
 514
 515         /* Outer loop uses 26 flops */
 516     }
 517
 518     /* Increment number of outer iterations */
 519     outeriter        += nri;
 520
 521     /* Update outer/inner flops */
 522
 523     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*131);
 524 }
 525 /*
 526  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_128_fma_double
 527  * Electrostatics interaction: Coulomb
 528  * VdW interaction:            LennardJones
 529  * Geometry:                   Water4-Particle
 530  * Calculate force/pot:        Force
 531  */
 532 void
 533 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_128_fma_double
 534                     (t_nblist                    * gmx_restrict       nlist,
 535                      rvec                        * gmx_restrict          xx,
 536                      rvec                        * gmx_restrict          ff,
 537                      struct t_forcerec           * gmx_restrict          fr,
 538                      t_mdatoms                   * gmx_restrict     mdatoms,
 539                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 540                      t_nrnb                      * gmx_restrict        nrnb)
 541 {
 542     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 543      * just 0 for non-waters.
 544      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 545      * jnr indices corresponding to data put in the four positions in the SIMD register.
 546      */
 547     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 548     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 549     int              jnrA,jnrB;
 550     int              j_coord_offsetA,j_coord_offsetB;
 551     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 552     real             rcutoff_scalar;
 553     real             *shiftvec,*fshift,*x,*f;
 554     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 555     int              vdwioffset0;
 556     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 557     int              vdwioffset1;
 558     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 559     int              vdwioffset2;
 560     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 561     int              vdwioffset3;
 562     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 563     int              vdwjidx0A,vdwjidx0B;
 564     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 565     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 566     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 567     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 568     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 569     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 570     real             *charge;
 571     int              nvdwtype;
 572     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 573     int              *vdwtype;
 574     real             *vdwparam;
 575     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 576     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 577     __m128d          dummy_mask,cutoff_mask;
 578     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 579     __m128d          one     = _mm_set1_pd(1.0);
 580     __m128d          two     = _mm_set1_pd(2.0);
 581     x                = xx[0];
 582     f                = ff[0];
 583
 584     nri              = nlist->nri;
 585     iinr             = nlist->iinr;
 586     jindex           = nlist->jindex;
 587     jjnr             = nlist->jjnr;
 588     shiftidx         = nlist->shift;
 589     gid              = nlist->gid;
 590     shiftvec         = fr->shift_vec[0];
 591     fshift           = fr->fshift[0];
 592     facel            = _mm_set1_pd(fr->ic->epsfac);
 593     charge           = mdatoms->chargeA;
 594     nvdwtype         = fr->ntype;
 595     vdwparam         = fr->nbfp;
 596     vdwtype          = mdatoms->typeA;
 597
 598     /* Setup water-specific parameters */
 599     inr              = nlist->iinr[0];
 600     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 601     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 602     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 603     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 604
 605     /* Avoid stupid compiler warnings */
 606     jnrA = jnrB = 0;
 607     j_coord_offsetA = 0;
 608     j_coord_offsetB = 0;
 609
 610     outeriter        = 0;
 611     inneriter        = 0;
 612
 613     /* Start outer loop over neighborlists */
 614     for(iidx=0; iidx<nri; iidx++)
 615     {
 616         /* Load shift vector for this list */
 617         i_shift_offset   = DIM*shiftidx[iidx];
 618
 619         /* Load limits for loop over neighbors */
 620         j_index_start    = jindex[iidx];
 621         j_index_end      = jindex[iidx+1];
 622
 623         /* Get outer coordinate index */
 624         inr              = iinr[iidx];
 625         i_coord_offset   = DIM*inr;
 626
 627         /* Load i particle coords and add shift vector */
 628         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 629                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 630
 631         fix0             = _mm_setzero_pd();
 632         fiy0             = _mm_setzero_pd();
 633         fiz0             = _mm_setzero_pd();
 634         fix1             = _mm_setzero_pd();
 635         fiy1             = _mm_setzero_pd();
 636         fiz1             = _mm_setzero_pd();
 637         fix2             = _mm_setzero_pd();
 638         fiy2             = _mm_setzero_pd();
 639         fiz2             = _mm_setzero_pd();
 640         fix3             = _mm_setzero_pd();
 641         fiy3             = _mm_setzero_pd();
 642         fiz3             = _mm_setzero_pd();
 643
 644         /* Start inner kernel loop */
 645         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 646         {
 647
 648             /* Get j neighbor index, and coordinate index */
 649             jnrA             = jjnr[jidx];
 650             jnrB             = jjnr[jidx+1];
 651             j_coord_offsetA  = DIM*jnrA;
 652             j_coord_offsetB  = DIM*jnrB;
 653
 654             /* load j atom coordinates */
 655             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 656                                               &jx0,&jy0,&jz0);
 657
 658             /* Calculate displacement vector */
 659             dx00             = _mm_sub_pd(ix0,jx0);
 660             dy00             = _mm_sub_pd(iy0,jy0);
 661             dz00             = _mm_sub_pd(iz0,jz0);
 662             dx10             = _mm_sub_pd(ix1,jx0);
 663             dy10             = _mm_sub_pd(iy1,jy0);
 664             dz10             = _mm_sub_pd(iz1,jz0);
 665             dx20             = _mm_sub_pd(ix2,jx0);
 666             dy20             = _mm_sub_pd(iy2,jy0);
 667             dz20             = _mm_sub_pd(iz2,jz0);
 668             dx30             = _mm_sub_pd(ix3,jx0);
 669             dy30             = _mm_sub_pd(iy3,jy0);
 670             dz30             = _mm_sub_pd(iz3,jz0);
 671
 672             /* Calculate squared distance and things based on it */
 673             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 674             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 675             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 676             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 677
 678             rinv10           = avx128fma_invsqrt_d(rsq10);
 679             rinv20           = avx128fma_invsqrt_d(rsq20);
 680             rinv30           = avx128fma_invsqrt_d(rsq30);
 681
 682             rinvsq00         = avx128fma_inv_d(rsq00);
 683             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 684             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 685             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 686
 687             /* Load parameters for j particles */
 688             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 689             vdwjidx0A        = 2*vdwtype[jnrA+0];
 690             vdwjidx0B        = 2*vdwtype[jnrB+0];
 691
 692             fjx0             = _mm_setzero_pd();
 693             fjy0             = _mm_setzero_pd();
 694             fjz0             = _mm_setzero_pd();
 695
 696             /**************************
 697              * CALCULATE INTERACTIONS *
 698              **************************/
 699
 700             /* Compute parameters for interactions between i and j atoms */
 701             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 702                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 703
 704             /* LENNARD-JONES DISPERSION/REPULSION */
 705
 706             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 707             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 708
 709             fscal            = fvdw;
 710
 711             /* Update vectorial force */
 712             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 713             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 714             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 715
 716             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 717             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 718             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             /* Compute parameters for interactions between i and j atoms */
 725             qq10             = _mm_mul_pd(iq1,jq0);
 726
 727             /* COULOMB ELECTROSTATICS */
 728             velec            = _mm_mul_pd(qq10,rinv10);
 729             felec            = _mm_mul_pd(velec,rinvsq10);
 730
 731             fscal            = felec;
 732
 733             /* Update vectorial force */
 734             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 735             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 736             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 737
 738             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 739             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 740             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 741
 742             /**************************
 743              * CALCULATE INTERACTIONS *
 744              **************************/
 745
 746             /* Compute parameters for interactions between i and j atoms */
 747             qq20             = _mm_mul_pd(iq2,jq0);
 748
 749             /* COULOMB ELECTROSTATICS */
 750             velec            = _mm_mul_pd(qq20,rinv20);
 751             felec            = _mm_mul_pd(velec,rinvsq20);
 752
 753             fscal            = felec;
 754
 755             /* Update vectorial force */
 756             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 757             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 758             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 759
 760             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 761             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 762             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             /* Compute parameters for interactions between i and j atoms */
 769             qq30             = _mm_mul_pd(iq3,jq0);
 770
 771             /* COULOMB ELECTROSTATICS */
 772             velec            = _mm_mul_pd(qq30,rinv30);
 773             felec            = _mm_mul_pd(velec,rinvsq30);
 774
 775             fscal            = felec;
 776
 777             /* Update vectorial force */
 778             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 779             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 780             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 781
 782             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 783             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 784             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 785
 786             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 787
 788             /* Inner loop uses 123 flops */
 789         }
 790
 791         if(jidx<j_index_end)
 792         {
 793
 794             jnrA             = jjnr[jidx];
 795             j_coord_offsetA  = DIM*jnrA;
 796
 797             /* load j atom coordinates */
 798             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 799                                               &jx0,&jy0,&jz0);
 800
 801             /* Calculate displacement vector */
 802             dx00             = _mm_sub_pd(ix0,jx0);
 803             dy00             = _mm_sub_pd(iy0,jy0);
 804             dz00             = _mm_sub_pd(iz0,jz0);
 805             dx10             = _mm_sub_pd(ix1,jx0);
 806             dy10             = _mm_sub_pd(iy1,jy0);
 807             dz10             = _mm_sub_pd(iz1,jz0);
 808             dx20             = _mm_sub_pd(ix2,jx0);
 809             dy20             = _mm_sub_pd(iy2,jy0);
 810             dz20             = _mm_sub_pd(iz2,jz0);
 811             dx30             = _mm_sub_pd(ix3,jx0);
 812             dy30             = _mm_sub_pd(iy3,jy0);
 813             dz30             = _mm_sub_pd(iz3,jz0);
 814
 815             /* Calculate squared distance and things based on it */
 816             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 817             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 818             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 819             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 820
 821             rinv10           = avx128fma_invsqrt_d(rsq10);
 822             rinv20           = avx128fma_invsqrt_d(rsq20);
 823             rinv30           = avx128fma_invsqrt_d(rsq30);
 824
 825             rinvsq00         = avx128fma_inv_d(rsq00);
 826             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 827             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 828             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 829
 830             /* Load parameters for j particles */
 831             jq0              = _mm_load_sd(charge+jnrA+0);
 832             vdwjidx0A        = 2*vdwtype[jnrA+0];
 833
 834             fjx0             = _mm_setzero_pd();
 835             fjy0             = _mm_setzero_pd();
 836             fjz0             = _mm_setzero_pd();
 837
 838             /**************************
 839              * CALCULATE INTERACTIONS *
 840              **************************/
 841
 842             /* Compute parameters for interactions between i and j atoms */
 843             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 844
 845             /* LENNARD-JONES DISPERSION/REPULSION */
 846
 847             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 848             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 849
 850             fscal            = fvdw;
 851
 852             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 853
 854             /* Update vectorial force */
 855             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 856             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 857             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 858
 859             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 860             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 861             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 862
 863             /**************************
 864              * CALCULATE INTERACTIONS *
 865              **************************/
 866
 867             /* Compute parameters for interactions between i and j atoms */
 868             qq10             = _mm_mul_pd(iq1,jq0);
 869
 870             /* COULOMB ELECTROSTATICS */
 871             velec            = _mm_mul_pd(qq10,rinv10);
 872             felec            = _mm_mul_pd(velec,rinvsq10);
 873
 874             fscal            = felec;
 875
 876             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 877
 878             /* Update vectorial force */
 879             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 880             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 881             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 882
 883             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 884             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 885             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq20             = _mm_mul_pd(iq2,jq0);
 893
 894             /* COULOMB ELECTROSTATICS */
 895             velec            = _mm_mul_pd(qq20,rinv20);
 896             felec            = _mm_mul_pd(velec,rinvsq20);
 897
 898             fscal            = felec;
 899
 900             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 901
 902             /* Update vectorial force */
 903             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 904             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 905             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 906
 907             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 908             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 909             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             /* Compute parameters for interactions between i and j atoms */
 916             qq30             = _mm_mul_pd(iq3,jq0);
 917
 918             /* COULOMB ELECTROSTATICS */
 919             velec            = _mm_mul_pd(qq30,rinv30);
 920             felec            = _mm_mul_pd(velec,rinvsq30);
 921
 922             fscal            = felec;
 923
 924             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 925
 926             /* Update vectorial force */
 927             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 928             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 929             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 930
 931             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 932             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 933             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 934
 935             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 936
 937             /* Inner loop uses 123 flops */
 938         }
 939
 940         /* End of innermost loop */
 941
 942         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 943                                               f+i_coord_offset,fshift+i_shift_offset);
 944
 945         /* Increment number of inner iterations */
 946         inneriter                  += j_index_end - j_index_start;
 947
 948         /* Outer loop uses 24 flops */
 949     }
 950
 951     /* Increment number of outer iterations */
 952     outeriter        += nri;
 953
 954     /* Update outer/inner flops */
 955
 956     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
 957 }