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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_avx_128_fma_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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128d          dummy_mask,cutoff_mask;
 105     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 106     __m128d          one     = _mm_set1_pd(1.0);
 107     __m128d          two     = _mm_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 128     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 129     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 130     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 156                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 157
 158         fix0             = _mm_setzero_pd();
 159         fiy0             = _mm_setzero_pd();
 160         fiz0             = _mm_setzero_pd();
 161         fix1             = _mm_setzero_pd();
 162         fiy1             = _mm_setzero_pd();
 163         fiz1             = _mm_setzero_pd();
 164         fix2             = _mm_setzero_pd();
 165         fiy2             = _mm_setzero_pd();
 166         fiz2             = _mm_setzero_pd();
 167         fix3             = _mm_setzero_pd();
 168         fiy3             = _mm_setzero_pd();
 169         fiz3             = _mm_setzero_pd();
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_pd();
 173         vvdwsum          = _mm_setzero_pd();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_pd(ix0,jx0);
 191             dy00             = _mm_sub_pd(iy0,jy0);
 192             dz00             = _mm_sub_pd(iz0,jz0);
 193             dx10             = _mm_sub_pd(ix1,jx0);
 194             dy10             = _mm_sub_pd(iy1,jy0);
 195             dz10             = _mm_sub_pd(iz1,jz0);
 196             dx20             = _mm_sub_pd(ix2,jx0);
 197             dy20             = _mm_sub_pd(iy2,jy0);
 198             dz20             = _mm_sub_pd(iz2,jz0);
 199             dx30             = _mm_sub_pd(ix3,jx0);
 200             dy30             = _mm_sub_pd(iy3,jy0);
 201             dz30             = _mm_sub_pd(iz3,jz0);
 202
 203             /* Calculate squared distance and things based on it */
 204             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 205             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 206             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 207             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 208
 209             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 210             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 211             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 212
 213             rinvsq00         = gmx_mm_inv_pd(rsq00);
 214             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 215             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 216             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 217
 218             /* Load parameters for j particles */
 219             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 220             vdwjidx0A        = 2*vdwtype[jnrA+0];
 221             vdwjidx0B        = 2*vdwtype[jnrB+0];
 222
 223             fjx0             = _mm_setzero_pd();
 224             fjy0             = _mm_setzero_pd();
 225             fjz0             = _mm_setzero_pd();
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             /* Compute parameters for interactions between i and j atoms */
 232             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 233                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 234
 235             /* LENNARD-JONES DISPERSION/REPULSION */
 236
 237             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 238             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 239             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 240             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 241             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 245
 246             fscal            = fvdw;
 247
 248             /* Update vectorial force */
 249             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 250             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 251             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 252
 253             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 254             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 255             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             qq10             = _mm_mul_pd(iq1,jq0);
 263
 264             /* COULOMB ELECTROSTATICS */
 265             velec            = _mm_mul_pd(qq10,rinv10);
 266             felec            = _mm_mul_pd(velec,rinvsq10);
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             velecsum         = _mm_add_pd(velecsum,velec);
 270
 271             fscal            = felec;
 272
 273             /* Update vectorial force */
 274             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 275             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 276             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 277
 278             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 279             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 280             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             qq20             = _mm_mul_pd(iq2,jq0);
 288
 289             /* COULOMB ELECTROSTATICS */
 290             velec            = _mm_mul_pd(qq20,rinv20);
 291             felec            = _mm_mul_pd(velec,rinvsq20);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velecsum         = _mm_add_pd(velecsum,velec);
 295
 296             fscal            = felec;
 297
 298             /* Update vectorial force */
 299             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 300             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 301             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 302
 303             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 304             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 305             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             qq30             = _mm_mul_pd(iq3,jq0);
 313
 314             /* COULOMB ELECTROSTATICS */
 315             velec            = _mm_mul_pd(qq30,rinv30);
 316             felec            = _mm_mul_pd(velec,rinvsq30);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velecsum         = _mm_add_pd(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             /* Update vectorial force */
 324             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 325             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 326             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 327
 328             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 329             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 330             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 331
 332             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 333
 334             /* Inner loop uses 131 flops */
 335         }
 336
 337         if(jidx<j_index_end)
 338         {
 339
 340             jnrA             = jjnr[jidx];
 341             j_coord_offsetA  = DIM*jnrA;
 342
 343             /* load j atom coordinates */
 344             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 345                                               &jx0,&jy0,&jz0);
 346
 347             /* Calculate displacement vector */
 348             dx00             = _mm_sub_pd(ix0,jx0);
 349             dy00             = _mm_sub_pd(iy0,jy0);
 350             dz00             = _mm_sub_pd(iz0,jz0);
 351             dx10             = _mm_sub_pd(ix1,jx0);
 352             dy10             = _mm_sub_pd(iy1,jy0);
 353             dz10             = _mm_sub_pd(iz1,jz0);
 354             dx20             = _mm_sub_pd(ix2,jx0);
 355             dy20             = _mm_sub_pd(iy2,jy0);
 356             dz20             = _mm_sub_pd(iz2,jz0);
 357             dx30             = _mm_sub_pd(ix3,jx0);
 358             dy30             = _mm_sub_pd(iy3,jy0);
 359             dz30             = _mm_sub_pd(iz3,jz0);
 360
 361             /* Calculate squared distance and things based on it */
 362             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 363             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 364             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 365             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 366
 367             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 368             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 369             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 370
 371             rinvsq00         = gmx_mm_inv_pd(rsq00);
 372             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 373             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 374             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 375
 376             /* Load parameters for j particles */
 377             jq0              = _mm_load_sd(charge+jnrA+0);
 378             vdwjidx0A        = 2*vdwtype[jnrA+0];
 379
 380             fjx0             = _mm_setzero_pd();
 381             fjy0             = _mm_setzero_pd();
 382             fjz0             = _mm_setzero_pd();
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             /* Compute parameters for interactions between i and j atoms */
 389             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 390
 391             /* LENNARD-JONES DISPERSION/REPULSION */
 392
 393             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 394             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 395             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 396             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 397             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 398
 399             /* Update potential sum for this i atom from the interaction with this j atom. */
 400             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 401             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 402
 403             fscal            = fvdw;
 404
 405             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 406
 407             /* Update vectorial force */
 408             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 409             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 410             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 411
 412             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 413             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 414             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 415
 416             /**************************
 417              * CALCULATE INTERACTIONS *
 418              **************************/
 419
 420             /* Compute parameters for interactions between i and j atoms */
 421             qq10             = _mm_mul_pd(iq1,jq0);
 422
 423             /* COULOMB ELECTROSTATICS */
 424             velec            = _mm_mul_pd(qq10,rinv10);
 425             felec            = _mm_mul_pd(velec,rinvsq10);
 426
 427             /* Update potential sum for this i atom from the interaction with this j atom. */
 428             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 429             velecsum         = _mm_add_pd(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 434
 435             /* Update vectorial force */
 436             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 437             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 438             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 439
 440             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 441             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 442             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             /* Compute parameters for interactions between i and j atoms */
 449             qq20             = _mm_mul_pd(iq2,jq0);
 450
 451             /* COULOMB ELECTROSTATICS */
 452             velec            = _mm_mul_pd(qq20,rinv20);
 453             felec            = _mm_mul_pd(velec,rinvsq20);
 454
 455             /* Update potential sum for this i atom from the interaction with this j atom. */
 456             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 457             velecsum         = _mm_add_pd(velecsum,velec);
 458
 459             fscal            = felec;
 460
 461             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 462
 463             /* Update vectorial force */
 464             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 465             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 466             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 467
 468             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 469             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 470             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             /* Compute parameters for interactions between i and j atoms */
 477             qq30             = _mm_mul_pd(iq3,jq0);
 478
 479             /* COULOMB ELECTROSTATICS */
 480             velec            = _mm_mul_pd(qq30,rinv30);
 481             felec            = _mm_mul_pd(velec,rinvsq30);
 482
 483             /* Update potential sum for this i atom from the interaction with this j atom. */
 484             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 485             velecsum         = _mm_add_pd(velecsum,velec);
 486
 487             fscal            = felec;
 488
 489             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 490
 491             /* Update vectorial force */
 492             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 493             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 494             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 495
 496             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 497             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 498             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 499
 500             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 501
 502             /* Inner loop uses 131 flops */
 503         }
 504
 505         /* End of innermost loop */
 506
 507         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 508                                               f+i_coord_offset,fshift+i_shift_offset);
 509
 510         ggid                        = gid[iidx];
 511         /* Update potential energies */
 512         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 513         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 514
 515         /* Increment number of inner iterations */
 516         inneriter                  += j_index_end - j_index_start;
 517
 518         /* Outer loop uses 26 flops */
 519     }
 520
 521     /* Increment number of outer iterations */
 522     outeriter        += nri;
 523
 524     /* Update outer/inner flops */
 525
 526     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*131);
 527 }
 528 /*
 529  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_128_fma_double
 530  * Electrostatics interaction: Coulomb
 531  * VdW interaction:            LennardJones
 532  * Geometry:                   Water4-Particle
 533  * Calculate force/pot:        Force
 534  */
 535 void
 536 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_128_fma_double
 537                     (t_nblist                    * gmx_restrict       nlist,
 538                      rvec                        * gmx_restrict          xx,
 539                      rvec                        * gmx_restrict          ff,
 540                      t_forcerec                  * gmx_restrict          fr,
 541                      t_mdatoms                   * gmx_restrict     mdatoms,
 542                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 543                      t_nrnb                      * gmx_restrict        nrnb)
 544 {
 545     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 546      * just 0 for non-waters.
 547      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 548      * jnr indices corresponding to data put in the four positions in the SIMD register.
 549      */
 550     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 551     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 552     int              jnrA,jnrB;
 553     int              j_coord_offsetA,j_coord_offsetB;
 554     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 555     real             rcutoff_scalar;
 556     real             *shiftvec,*fshift,*x,*f;
 557     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 558     int              vdwioffset0;
 559     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 560     int              vdwioffset1;
 561     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 562     int              vdwioffset2;
 563     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 564     int              vdwioffset3;
 565     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 566     int              vdwjidx0A,vdwjidx0B;
 567     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 568     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 569     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 570     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 571     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 572     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 573     real             *charge;
 574     int              nvdwtype;
 575     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 576     int              *vdwtype;
 577     real             *vdwparam;
 578     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 579     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 580     __m128d          dummy_mask,cutoff_mask;
 581     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 582     __m128d          one     = _mm_set1_pd(1.0);
 583     __m128d          two     = _mm_set1_pd(2.0);
 584     x                = xx[0];
 585     f                = ff[0];
 586
 587     nri              = nlist->nri;
 588     iinr             = nlist->iinr;
 589     jindex           = nlist->jindex;
 590     jjnr             = nlist->jjnr;
 591     shiftidx         = nlist->shift;
 592     gid              = nlist->gid;
 593     shiftvec         = fr->shift_vec[0];
 594     fshift           = fr->fshift[0];
 595     facel            = _mm_set1_pd(fr->epsfac);
 596     charge           = mdatoms->chargeA;
 597     nvdwtype         = fr->ntype;
 598     vdwparam         = fr->nbfp;
 599     vdwtype          = mdatoms->typeA;
 600
 601     /* Setup water-specific parameters */
 602     inr              = nlist->iinr[0];
 603     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 604     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 605     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 606     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 607
 608     /* Avoid stupid compiler warnings */
 609     jnrA = jnrB = 0;
 610     j_coord_offsetA = 0;
 611     j_coord_offsetB = 0;
 612
 613     outeriter        = 0;
 614     inneriter        = 0;
 615
 616     /* Start outer loop over neighborlists */
 617     for(iidx=0; iidx<nri; iidx++)
 618     {
 619         /* Load shift vector for this list */
 620         i_shift_offset   = DIM*shiftidx[iidx];
 621
 622         /* Load limits for loop over neighbors */
 623         j_index_start    = jindex[iidx];
 624         j_index_end      = jindex[iidx+1];
 625
 626         /* Get outer coordinate index */
 627         inr              = iinr[iidx];
 628         i_coord_offset   = DIM*inr;
 629
 630         /* Load i particle coords and add shift vector */
 631         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 632                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 633
 634         fix0             = _mm_setzero_pd();
 635         fiy0             = _mm_setzero_pd();
 636         fiz0             = _mm_setzero_pd();
 637         fix1             = _mm_setzero_pd();
 638         fiy1             = _mm_setzero_pd();
 639         fiz1             = _mm_setzero_pd();
 640         fix2             = _mm_setzero_pd();
 641         fiy2             = _mm_setzero_pd();
 642         fiz2             = _mm_setzero_pd();
 643         fix3             = _mm_setzero_pd();
 644         fiy3             = _mm_setzero_pd();
 645         fiz3             = _mm_setzero_pd();
 646
 647         /* Start inner kernel loop */
 648         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 649         {
 650
 651             /* Get j neighbor index, and coordinate index */
 652             jnrA             = jjnr[jidx];
 653             jnrB             = jjnr[jidx+1];
 654             j_coord_offsetA  = DIM*jnrA;
 655             j_coord_offsetB  = DIM*jnrB;
 656
 657             /* load j atom coordinates */
 658             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 659                                               &jx0,&jy0,&jz0);
 660
 661             /* Calculate displacement vector */
 662             dx00             = _mm_sub_pd(ix0,jx0);
 663             dy00             = _mm_sub_pd(iy0,jy0);
 664             dz00             = _mm_sub_pd(iz0,jz0);
 665             dx10             = _mm_sub_pd(ix1,jx0);
 666             dy10             = _mm_sub_pd(iy1,jy0);
 667             dz10             = _mm_sub_pd(iz1,jz0);
 668             dx20             = _mm_sub_pd(ix2,jx0);
 669             dy20             = _mm_sub_pd(iy2,jy0);
 670             dz20             = _mm_sub_pd(iz2,jz0);
 671             dx30             = _mm_sub_pd(ix3,jx0);
 672             dy30             = _mm_sub_pd(iy3,jy0);
 673             dz30             = _mm_sub_pd(iz3,jz0);
 674
 675             /* Calculate squared distance and things based on it */
 676             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 677             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 678             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 679             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 680
 681             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 682             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 683             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 684
 685             rinvsq00         = gmx_mm_inv_pd(rsq00);
 686             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 687             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 688             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 689
 690             /* Load parameters for j particles */
 691             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 692             vdwjidx0A        = 2*vdwtype[jnrA+0];
 693             vdwjidx0B        = 2*vdwtype[jnrB+0];
 694
 695             fjx0             = _mm_setzero_pd();
 696             fjy0             = _mm_setzero_pd();
 697             fjz0             = _mm_setzero_pd();
 698
 699             /**************************
 700              * CALCULATE INTERACTIONS *
 701              **************************/
 702
 703             /* Compute parameters for interactions between i and j atoms */
 704             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 705                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 706
 707             /* LENNARD-JONES DISPERSION/REPULSION */
 708
 709             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 710             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 711
 712             fscal            = fvdw;
 713
 714             /* Update vectorial force */
 715             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 716             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 717             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 718
 719             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 720             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 721             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 722
 723             /**************************
 724              * CALCULATE INTERACTIONS *
 725              **************************/
 726
 727             /* Compute parameters for interactions between i and j atoms */
 728             qq10             = _mm_mul_pd(iq1,jq0);
 729
 730             /* COULOMB ELECTROSTATICS */
 731             velec            = _mm_mul_pd(qq10,rinv10);
 732             felec            = _mm_mul_pd(velec,rinvsq10);
 733
 734             fscal            = felec;
 735
 736             /* Update vectorial force */
 737             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 738             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 739             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 740
 741             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 742             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 743             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 744
 745             /**************************
 746              * CALCULATE INTERACTIONS *
 747              **************************/
 748
 749             /* Compute parameters for interactions between i and j atoms */
 750             qq20             = _mm_mul_pd(iq2,jq0);
 751
 752             /* COULOMB ELECTROSTATICS */
 753             velec            = _mm_mul_pd(qq20,rinv20);
 754             felec            = _mm_mul_pd(velec,rinvsq20);
 755
 756             fscal            = felec;
 757
 758             /* Update vectorial force */
 759             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 760             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 761             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 762
 763             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 764             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 765             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 766
 767             /**************************
 768              * CALCULATE INTERACTIONS *
 769              **************************/
 770
 771             /* Compute parameters for interactions between i and j atoms */
 772             qq30             = _mm_mul_pd(iq3,jq0);
 773
 774             /* COULOMB ELECTROSTATICS */
 775             velec            = _mm_mul_pd(qq30,rinv30);
 776             felec            = _mm_mul_pd(velec,rinvsq30);
 777
 778             fscal            = felec;
 779
 780             /* Update vectorial force */
 781             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 782             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 783             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 784
 785             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 786             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 787             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 788
 789             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 790
 791             /* Inner loop uses 123 flops */
 792         }
 793
 794         if(jidx<j_index_end)
 795         {
 796
 797             jnrA             = jjnr[jidx];
 798             j_coord_offsetA  = DIM*jnrA;
 799
 800             /* load j atom coordinates */
 801             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 802                                               &jx0,&jy0,&jz0);
 803
 804             /* Calculate displacement vector */
 805             dx00             = _mm_sub_pd(ix0,jx0);
 806             dy00             = _mm_sub_pd(iy0,jy0);
 807             dz00             = _mm_sub_pd(iz0,jz0);
 808             dx10             = _mm_sub_pd(ix1,jx0);
 809             dy10             = _mm_sub_pd(iy1,jy0);
 810             dz10             = _mm_sub_pd(iz1,jz0);
 811             dx20             = _mm_sub_pd(ix2,jx0);
 812             dy20             = _mm_sub_pd(iy2,jy0);
 813             dz20             = _mm_sub_pd(iz2,jz0);
 814             dx30             = _mm_sub_pd(ix3,jx0);
 815             dy30             = _mm_sub_pd(iy3,jy0);
 816             dz30             = _mm_sub_pd(iz3,jz0);
 817
 818             /* Calculate squared distance and things based on it */
 819             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 820             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 821             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 822             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 823
 824             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 825             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 826             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 827
 828             rinvsq00         = gmx_mm_inv_pd(rsq00);
 829             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 830             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 831             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 832
 833             /* Load parameters for j particles */
 834             jq0              = _mm_load_sd(charge+jnrA+0);
 835             vdwjidx0A        = 2*vdwtype[jnrA+0];
 836
 837             fjx0             = _mm_setzero_pd();
 838             fjy0             = _mm_setzero_pd();
 839             fjz0             = _mm_setzero_pd();
 840
 841             /**************************
 842              * CALCULATE INTERACTIONS *
 843              **************************/
 844
 845             /* Compute parameters for interactions between i and j atoms */
 846             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 847
 848             /* LENNARD-JONES DISPERSION/REPULSION */
 849
 850             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 851             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 852
 853             fscal            = fvdw;
 854
 855             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 856
 857             /* Update vectorial force */
 858             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 859             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 860             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 861
 862             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 863             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 864             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq10             = _mm_mul_pd(iq1,jq0);
 872
 873             /* COULOMB ELECTROSTATICS */
 874             velec            = _mm_mul_pd(qq10,rinv10);
 875             felec            = _mm_mul_pd(velec,rinvsq10);
 876
 877             fscal            = felec;
 878
 879             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 880
 881             /* Update vectorial force */
 882             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 883             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 884             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 885
 886             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 887             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 888             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 889
 890             /**************************
 891              * CALCULATE INTERACTIONS *
 892              **************************/
 893
 894             /* Compute parameters for interactions between i and j atoms */
 895             qq20             = _mm_mul_pd(iq2,jq0);
 896
 897             /* COULOMB ELECTROSTATICS */
 898             velec            = _mm_mul_pd(qq20,rinv20);
 899             felec            = _mm_mul_pd(velec,rinvsq20);
 900
 901             fscal            = felec;
 902
 903             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 904
 905             /* Update vectorial force */
 906             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 907             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 908             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 909
 910             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 911             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 912             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             /* Compute parameters for interactions between i and j atoms */
 919             qq30             = _mm_mul_pd(iq3,jq0);
 920
 921             /* COULOMB ELECTROSTATICS */
 922             velec            = _mm_mul_pd(qq30,rinv30);
 923             felec            = _mm_mul_pd(velec,rinvsq30);
 924
 925             fscal            = felec;
 926
 927             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 928
 929             /* Update vectorial force */
 930             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 931             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 932             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 933
 934             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 935             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 936             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 937
 938             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 939
 940             /* Inner loop uses 123 flops */
 941         }
 942
 943         /* End of innermost loop */
 944
 945         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 946                                               f+i_coord_offset,fshift+i_shift_offset);
 947
 948         /* Increment number of inner iterations */
 949         inneriter                  += j_index_end - j_index_start;
 950
 951         /* Outer loop uses 24 flops */
 952     }
 953
 954     /* Increment number of outer iterations */
 955     outeriter        += nri;
 956
 957     /* Update outer/inner flops */
 958
 959     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
 960 }