src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128d          dummy_mask,cutoff_mask;
 100     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 101     __m128d          one     = _mm_set1_pd(1.0);
 102     __m128d          two     = _mm_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 123     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 124     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 125     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 151                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 152
 153         fix0             = _mm_setzero_pd();
 154         fiy0             = _mm_setzero_pd();
 155         fiz0             = _mm_setzero_pd();
 156         fix1             = _mm_setzero_pd();
 157         fiy1             = _mm_setzero_pd();
 158         fiz1             = _mm_setzero_pd();
 159         fix2             = _mm_setzero_pd();
 160         fiy2             = _mm_setzero_pd();
 161         fiz2             = _mm_setzero_pd();
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_pd();
 165         vvdwsum          = _mm_setzero_pd();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm_sub_pd(ix0,jx0);
 183             dy00             = _mm_sub_pd(iy0,jy0);
 184             dz00             = _mm_sub_pd(iz0,jz0);
 185             dx10             = _mm_sub_pd(ix1,jx0);
 186             dy10             = _mm_sub_pd(iy1,jy0);
 187             dz10             = _mm_sub_pd(iz1,jz0);
 188             dx20             = _mm_sub_pd(ix2,jx0);
 189             dy20             = _mm_sub_pd(iy2,jy0);
 190             dz20             = _mm_sub_pd(iz2,jz0);
 191
 192             /* Calculate squared distance and things based on it */
 193             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 194             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 195             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 196
 197             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 198             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 199             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 200
 201             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 202             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 203             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 204
 205             /* Load parameters for j particles */
 206             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 207             vdwjidx0A        = 2*vdwtype[jnrA+0];
 208             vdwjidx0B        = 2*vdwtype[jnrB+0];
 209
 210             fjx0             = _mm_setzero_pd();
 211             fjy0             = _mm_setzero_pd();
 212             fjz0             = _mm_setzero_pd();
 213
 214             /**************************
 215              * CALCULATE INTERACTIONS *
 216              **************************/
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _mm_mul_pd(iq0,jq0);
 220             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 221                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 222
 223             /* COULOMB ELECTROSTATICS */
 224             velec            = _mm_mul_pd(qq00,rinv00);
 225             felec            = _mm_mul_pd(velec,rinvsq00);
 226
 227             /* LENNARD-JONES DISPERSION/REPULSION */
 228
 229             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 230             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 231             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 232             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 233             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 234
 235             /* Update potential sum for this i atom from the interaction with this j atom. */
 236             velecsum         = _mm_add_pd(velecsum,velec);
 237             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 238
 239             fscal            = _mm_add_pd(felec,fvdw);
 240
 241             /* Update vectorial force */
 242             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 243             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 244             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 245
 246             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 247             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 248             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             qq10             = _mm_mul_pd(iq1,jq0);
 256
 257             /* COULOMB ELECTROSTATICS */
 258             velec            = _mm_mul_pd(qq10,rinv10);
 259             felec            = _mm_mul_pd(velec,rinvsq10);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velecsum         = _mm_add_pd(velecsum,velec);
 263
 264             fscal            = felec;
 265
 266             /* Update vectorial force */
 267             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 268             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 269             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 270
 271             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 272             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 273             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq20             = _mm_mul_pd(iq2,jq0);
 281
 282             /* COULOMB ELECTROSTATICS */
 283             velec            = _mm_mul_pd(qq20,rinv20);
 284             felec            = _mm_mul_pd(velec,rinvsq20);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velecsum         = _mm_add_pd(velecsum,velec);
 288
 289             fscal            = felec;
 290
 291             /* Update vectorial force */
 292             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 293             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 294             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 295
 296             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 297             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 298             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 299
 300             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 301
 302             /* Inner loop uses 108 flops */
 303         }
 304
 305         if(jidx<j_index_end)
 306         {
 307
 308             jnrA             = jjnr[jidx];
 309             j_coord_offsetA  = DIM*jnrA;
 310
 311             /* load j atom coordinates */
 312             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 313                                               &jx0,&jy0,&jz0);
 314
 315             /* Calculate displacement vector */
 316             dx00             = _mm_sub_pd(ix0,jx0);
 317             dy00             = _mm_sub_pd(iy0,jy0);
 318             dz00             = _mm_sub_pd(iz0,jz0);
 319             dx10             = _mm_sub_pd(ix1,jx0);
 320             dy10             = _mm_sub_pd(iy1,jy0);
 321             dz10             = _mm_sub_pd(iz1,jz0);
 322             dx20             = _mm_sub_pd(ix2,jx0);
 323             dy20             = _mm_sub_pd(iy2,jy0);
 324             dz20             = _mm_sub_pd(iz2,jz0);
 325
 326             /* Calculate squared distance and things based on it */
 327             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 328             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 329             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 330
 331             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 332             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 333             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 334
 335             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 336             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 337             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 338
 339             /* Load parameters for j particles */
 340             jq0              = _mm_load_sd(charge+jnrA+0);
 341             vdwjidx0A        = 2*vdwtype[jnrA+0];
 342
 343             fjx0             = _mm_setzero_pd();
 344             fjy0             = _mm_setzero_pd();
 345             fjz0             = _mm_setzero_pd();
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq00             = _mm_mul_pd(iq0,jq0);
 353             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 354
 355             /* COULOMB ELECTROSTATICS */
 356             velec            = _mm_mul_pd(qq00,rinv00);
 357             felec            = _mm_mul_pd(velec,rinvsq00);
 358
 359             /* LENNARD-JONES DISPERSION/REPULSION */
 360
 361             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 362             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 363             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 364             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 365             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 369             velecsum         = _mm_add_pd(velecsum,velec);
 370             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 371             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 372
 373             fscal            = _mm_add_pd(felec,fvdw);
 374
 375             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 376
 377             /* Update vectorial force */
 378             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 379             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 380             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 381
 382             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 383             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 384             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 385
 386             /**************************
 387              * CALCULATE INTERACTIONS *
 388              **************************/
 389
 390             /* Compute parameters for interactions between i and j atoms */
 391             qq10             = _mm_mul_pd(iq1,jq0);
 392
 393             /* COULOMB ELECTROSTATICS */
 394             velec            = _mm_mul_pd(qq10,rinv10);
 395             felec            = _mm_mul_pd(velec,rinvsq10);
 396
 397             /* Update potential sum for this i atom from the interaction with this j atom. */
 398             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 399             velecsum         = _mm_add_pd(velecsum,velec);
 400
 401             fscal            = felec;
 402
 403             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 404
 405             /* Update vectorial force */
 406             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 407             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 408             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 409
 410             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 411             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 412             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 413
 414             /**************************
 415              * CALCULATE INTERACTIONS *
 416              **************************/
 417
 418             /* Compute parameters for interactions between i and j atoms */
 419             qq20             = _mm_mul_pd(iq2,jq0);
 420
 421             /* COULOMB ELECTROSTATICS */
 422             velec            = _mm_mul_pd(qq20,rinv20);
 423             felec            = _mm_mul_pd(velec,rinvsq20);
 424
 425             /* Update potential sum for this i atom from the interaction with this j atom. */
 426             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 427             velecsum         = _mm_add_pd(velecsum,velec);
 428
 429             fscal            = felec;
 430
 431             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 432
 433             /* Update vectorial force */
 434             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 435             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 436             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 437
 438             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 439             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 440             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 441
 442             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 443
 444             /* Inner loop uses 108 flops */
 445         }
 446
 447         /* End of innermost loop */
 448
 449         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 450                                               f+i_coord_offset,fshift+i_shift_offset);
 451
 452         ggid                        = gid[iidx];
 453         /* Update potential energies */
 454         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 455         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 456
 457         /* Increment number of inner iterations */
 458         inneriter                  += j_index_end - j_index_start;
 459
 460         /* Outer loop uses 20 flops */
 461     }
 462
 463     /* Increment number of outer iterations */
 464     outeriter        += nri;
 465
 466     /* Update outer/inner flops */
 467
 468     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*108);
 469 }
 470 /*
 471  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_double
 472  * Electrostatics interaction: Coulomb
 473  * VdW interaction:            LennardJones
 474  * Geometry:                   Water3-Particle
 475  * Calculate force/pot:        Force
 476  */
 477 void
 478 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_double
 479                     (t_nblist                    * gmx_restrict       nlist,
 480                      rvec                        * gmx_restrict          xx,
 481                      rvec                        * gmx_restrict          ff,
 482                      t_forcerec                  * gmx_restrict          fr,
 483                      t_mdatoms                   * gmx_restrict     mdatoms,
 484                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 485                      t_nrnb                      * gmx_restrict        nrnb)
 486 {
 487     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 488      * just 0 for non-waters.
 489      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 490      * jnr indices corresponding to data put in the four positions in the SIMD register.
 491      */
 492     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 493     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 494     int              jnrA,jnrB;
 495     int              j_coord_offsetA,j_coord_offsetB;
 496     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 497     real             rcutoff_scalar;
 498     real             *shiftvec,*fshift,*x,*f;
 499     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 500     int              vdwioffset0;
 501     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 502     int              vdwioffset1;
 503     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 504     int              vdwioffset2;
 505     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 506     int              vdwjidx0A,vdwjidx0B;
 507     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 508     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 509     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 510     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 511     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 512     real             *charge;
 513     int              nvdwtype;
 514     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 515     int              *vdwtype;
 516     real             *vdwparam;
 517     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 518     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 519     __m128d          dummy_mask,cutoff_mask;
 520     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 521     __m128d          one     = _mm_set1_pd(1.0);
 522     __m128d          two     = _mm_set1_pd(2.0);
 523     x                = xx[0];
 524     f                = ff[0];
 525
 526     nri              = nlist->nri;
 527     iinr             = nlist->iinr;
 528     jindex           = nlist->jindex;
 529     jjnr             = nlist->jjnr;
 530     shiftidx         = nlist->shift;
 531     gid              = nlist->gid;
 532     shiftvec         = fr->shift_vec[0];
 533     fshift           = fr->fshift[0];
 534     facel            = _mm_set1_pd(fr->epsfac);
 535     charge           = mdatoms->chargeA;
 536     nvdwtype         = fr->ntype;
 537     vdwparam         = fr->nbfp;
 538     vdwtype          = mdatoms->typeA;
 539
 540     /* Setup water-specific parameters */
 541     inr              = nlist->iinr[0];
 542     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 543     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 544     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 545     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 546
 547     /* Avoid stupid compiler warnings */
 548     jnrA = jnrB = 0;
 549     j_coord_offsetA = 0;
 550     j_coord_offsetB = 0;
 551
 552     outeriter        = 0;
 553     inneriter        = 0;
 554
 555     /* Start outer loop over neighborlists */
 556     for(iidx=0; iidx<nri; iidx++)
 557     {
 558         /* Load shift vector for this list */
 559         i_shift_offset   = DIM*shiftidx[iidx];
 560
 561         /* Load limits for loop over neighbors */
 562         j_index_start    = jindex[iidx];
 563         j_index_end      = jindex[iidx+1];
 564
 565         /* Get outer coordinate index */
 566         inr              = iinr[iidx];
 567         i_coord_offset   = DIM*inr;
 568
 569         /* Load i particle coords and add shift vector */
 570         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 571                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 572
 573         fix0             = _mm_setzero_pd();
 574         fiy0             = _mm_setzero_pd();
 575         fiz0             = _mm_setzero_pd();
 576         fix1             = _mm_setzero_pd();
 577         fiy1             = _mm_setzero_pd();
 578         fiz1             = _mm_setzero_pd();
 579         fix2             = _mm_setzero_pd();
 580         fiy2             = _mm_setzero_pd();
 581         fiz2             = _mm_setzero_pd();
 582
 583         /* Start inner kernel loop */
 584         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 585         {
 586
 587             /* Get j neighbor index, and coordinate index */
 588             jnrA             = jjnr[jidx];
 589             jnrB             = jjnr[jidx+1];
 590             j_coord_offsetA  = DIM*jnrA;
 591             j_coord_offsetB  = DIM*jnrB;
 592
 593             /* load j atom coordinates */
 594             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 595                                               &jx0,&jy0,&jz0);
 596
 597             /* Calculate displacement vector */
 598             dx00             = _mm_sub_pd(ix0,jx0);
 599             dy00             = _mm_sub_pd(iy0,jy0);
 600             dz00             = _mm_sub_pd(iz0,jz0);
 601             dx10             = _mm_sub_pd(ix1,jx0);
 602             dy10             = _mm_sub_pd(iy1,jy0);
 603             dz10             = _mm_sub_pd(iz1,jz0);
 604             dx20             = _mm_sub_pd(ix2,jx0);
 605             dy20             = _mm_sub_pd(iy2,jy0);
 606             dz20             = _mm_sub_pd(iz2,jz0);
 607
 608             /* Calculate squared distance and things based on it */
 609             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 610             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 611             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 612
 613             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 614             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 615             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 616
 617             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 618             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 619             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 620
 621             /* Load parameters for j particles */
 622             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 623             vdwjidx0A        = 2*vdwtype[jnrA+0];
 624             vdwjidx0B        = 2*vdwtype[jnrB+0];
 625
 626             fjx0             = _mm_setzero_pd();
 627             fjy0             = _mm_setzero_pd();
 628             fjz0             = _mm_setzero_pd();
 629
 630             /**************************
 631              * CALCULATE INTERACTIONS *
 632              **************************/
 633
 634             /* Compute parameters for interactions between i and j atoms */
 635             qq00             = _mm_mul_pd(iq0,jq0);
 636             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 637                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 638
 639             /* COULOMB ELECTROSTATICS */
 640             velec            = _mm_mul_pd(qq00,rinv00);
 641             felec            = _mm_mul_pd(velec,rinvsq00);
 642
 643             /* LENNARD-JONES DISPERSION/REPULSION */
 644
 645             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 646             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 647
 648             fscal            = _mm_add_pd(felec,fvdw);
 649
 650             /* Update vectorial force */
 651             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 652             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 653             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 654
 655             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 656             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 657             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             /* Compute parameters for interactions between i and j atoms */
 664             qq10             = _mm_mul_pd(iq1,jq0);
 665
 666             /* COULOMB ELECTROSTATICS */
 667             velec            = _mm_mul_pd(qq10,rinv10);
 668             felec            = _mm_mul_pd(velec,rinvsq10);
 669
 670             fscal            = felec;
 671
 672             /* Update vectorial force */
 673             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 674             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 675             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 676
 677             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 678             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 679             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 680
 681             /**************************
 682              * CALCULATE INTERACTIONS *
 683              **************************/
 684
 685             /* Compute parameters for interactions between i and j atoms */
 686             qq20             = _mm_mul_pd(iq2,jq0);
 687
 688             /* COULOMB ELECTROSTATICS */
 689             velec            = _mm_mul_pd(qq20,rinv20);
 690             felec            = _mm_mul_pd(velec,rinvsq20);
 691
 692             fscal            = felec;
 693
 694             /* Update vectorial force */
 695             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 696             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 697             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 698
 699             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 700             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 701             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 702
 703             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 704
 705             /* Inner loop uses 100 flops */
 706         }
 707
 708         if(jidx<j_index_end)
 709         {
 710
 711             jnrA             = jjnr[jidx];
 712             j_coord_offsetA  = DIM*jnrA;
 713
 714             /* load j atom coordinates */
 715             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 716                                               &jx0,&jy0,&jz0);
 717
 718             /* Calculate displacement vector */
 719             dx00             = _mm_sub_pd(ix0,jx0);
 720             dy00             = _mm_sub_pd(iy0,jy0);
 721             dz00             = _mm_sub_pd(iz0,jz0);
 722             dx10             = _mm_sub_pd(ix1,jx0);
 723             dy10             = _mm_sub_pd(iy1,jy0);
 724             dz10             = _mm_sub_pd(iz1,jz0);
 725             dx20             = _mm_sub_pd(ix2,jx0);
 726             dy20             = _mm_sub_pd(iy2,jy0);
 727             dz20             = _mm_sub_pd(iz2,jz0);
 728
 729             /* Calculate squared distance and things based on it */
 730             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 731             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 732             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 733
 734             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 735             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 736             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 737
 738             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 739             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 740             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 741
 742             /* Load parameters for j particles */
 743             jq0              = _mm_load_sd(charge+jnrA+0);
 744             vdwjidx0A        = 2*vdwtype[jnrA+0];
 745
 746             fjx0             = _mm_setzero_pd();
 747             fjy0             = _mm_setzero_pd();
 748             fjz0             = _mm_setzero_pd();
 749
 750             /**************************
 751              * CALCULATE INTERACTIONS *
 752              **************************/
 753
 754             /* Compute parameters for interactions between i and j atoms */
 755             qq00             = _mm_mul_pd(iq0,jq0);
 756             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 757
 758             /* COULOMB ELECTROSTATICS */
 759             velec            = _mm_mul_pd(qq00,rinv00);
 760             felec            = _mm_mul_pd(velec,rinvsq00);
 761
 762             /* LENNARD-JONES DISPERSION/REPULSION */
 763
 764             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 765             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 766
 767             fscal            = _mm_add_pd(felec,fvdw);
 768
 769             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 770
 771             /* Update vectorial force */
 772             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 773             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 774             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 775
 776             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 777             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 778             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 779
 780             /**************************
 781              * CALCULATE INTERACTIONS *
 782              **************************/
 783
 784             /* Compute parameters for interactions between i and j atoms */
 785             qq10             = _mm_mul_pd(iq1,jq0);
 786
 787             /* COULOMB ELECTROSTATICS */
 788             velec            = _mm_mul_pd(qq10,rinv10);
 789             felec            = _mm_mul_pd(velec,rinvsq10);
 790
 791             fscal            = felec;
 792
 793             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 794
 795             /* Update vectorial force */
 796             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 797             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 798             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 799
 800             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 801             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 802             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 803
 804             /**************************
 805              * CALCULATE INTERACTIONS *
 806              **************************/
 807
 808             /* Compute parameters for interactions between i and j atoms */
 809             qq20             = _mm_mul_pd(iq2,jq0);
 810
 811             /* COULOMB ELECTROSTATICS */
 812             velec            = _mm_mul_pd(qq20,rinv20);
 813             felec            = _mm_mul_pd(velec,rinvsq20);
 814
 815             fscal            = felec;
 816
 817             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 818
 819             /* Update vectorial force */
 820             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 821             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 822             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 823
 824             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 825             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 826             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 827
 828             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 829
 830             /* Inner loop uses 100 flops */
 831         }
 832
 833         /* End of innermost loop */
 834
 835         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 836                                               f+i_coord_offset,fshift+i_shift_offset);
 837
 838         /* Increment number of inner iterations */
 839         inneriter                  += j_index_end - j_index_start;
 840
 841         /* Outer loop uses 18 flops */
 842     }
 843
 844     /* Increment number of outer iterations */
 845     outeriter        += nri;
 846
 847     /* Update outer/inner flops */
 848
 849     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
 850 }