src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwLJSw_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 #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_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_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              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 103     real             rswitch_scalar,d_scalar;
 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     krf              = _mm_set1_pd(fr->ic->k_rf);
 122     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 123     crf              = _mm_set1_pd(fr->ic->c_rf);
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 131     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 132     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 133     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 136     rcutoff_scalar   = fr->rcoulomb;
 137     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 138     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 139
 140     rswitch_scalar   = fr->rvdw_switch;
 141     rswitch          = _mm_set1_pd(rswitch_scalar);
 142     /* Setup switch parameters */
 143     d_scalar         = rcutoff_scalar-rswitch_scalar;
 144     d                = _mm_set1_pd(d_scalar);
 145     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 146     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 147     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 148     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 149     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 150     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 177
 178         fix0             = _mm_setzero_pd();
 179         fiy0             = _mm_setzero_pd();
 180         fiz0             = _mm_setzero_pd();
 181         fix1             = _mm_setzero_pd();
 182         fiy1             = _mm_setzero_pd();
 183         fiz1             = _mm_setzero_pd();
 184         fix2             = _mm_setzero_pd();
 185         fiy2             = _mm_setzero_pd();
 186         fiz2             = _mm_setzero_pd();
 187         fix3             = _mm_setzero_pd();
 188         fiy3             = _mm_setzero_pd();
 189         fiz3             = _mm_setzero_pd();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_pd();
 193         vvdwsum          = _mm_setzero_pd();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204
 205             /* load j atom coordinates */
 206             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_pd(ix0,jx0);
 211             dy00             = _mm_sub_pd(iy0,jy0);
 212             dz00             = _mm_sub_pd(iz0,jz0);
 213             dx10             = _mm_sub_pd(ix1,jx0);
 214             dy10             = _mm_sub_pd(iy1,jy0);
 215             dz10             = _mm_sub_pd(iz1,jz0);
 216             dx20             = _mm_sub_pd(ix2,jx0);
 217             dy20             = _mm_sub_pd(iy2,jy0);
 218             dz20             = _mm_sub_pd(iz2,jz0);
 219             dx30             = _mm_sub_pd(ix3,jx0);
 220             dy30             = _mm_sub_pd(iy3,jy0);
 221             dz30             = _mm_sub_pd(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 225             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 226             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 227             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 228
 229             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 230             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 231             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 232             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 233
 234             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 235             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 236             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 237             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243
 244             fjx0             = _mm_setzero_pd();
 245             fjy0             = _mm_setzero_pd();
 246             fjz0             = _mm_setzero_pd();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             if (gmx_mm_any_lt(rsq00,rcutoff2))
 253             {
 254
 255             r00              = _mm_mul_pd(rsq00,rinv00);
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 259                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 260
 261             /* LENNARD-JONES DISPERSION/REPULSION */
 262
 263             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 264             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 265             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 266             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 267             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 268
 269             d                = _mm_sub_pd(r00,rswitch);
 270             d                = _mm_max_pd(d,_mm_setzero_pd());
 271             d2               = _mm_mul_pd(d,d);
 272             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 273
 274             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 275
 276             /* Evaluate switch function */
 277             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 278             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 279             vvdw             = _mm_mul_pd(vvdw,sw);
 280             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 284             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 285
 286             fscal            = fvdw;
 287
 288             fscal            = _mm_and_pd(fscal,cutoff_mask);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 292             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 293             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 294
 295             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 296             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 297             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 298
 299             }
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             if (gmx_mm_any_lt(rsq10,rcutoff2))
 306             {
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq10             = _mm_mul_pd(iq1,jq0);
 310
 311             /* REACTION-FIELD ELECTROSTATICS */
 312             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 313             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 314
 315             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velec            = _mm_and_pd(velec,cutoff_mask);
 319             velecsum         = _mm_add_pd(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             fscal            = _mm_and_pd(fscal,cutoff_mask);
 324
 325             /* Update vectorial force */
 326             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 327             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 328             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 329
 330             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 331             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 332             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 333
 334             }
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             if (gmx_mm_any_lt(rsq20,rcutoff2))
 341             {
 342
 343             /* Compute parameters for interactions between i and j atoms */
 344             qq20             = _mm_mul_pd(iq2,jq0);
 345
 346             /* REACTION-FIELD ELECTROSTATICS */
 347             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 348             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 349
 350             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velec            = _mm_and_pd(velec,cutoff_mask);
 354             velecsum         = _mm_add_pd(velecsum,velec);
 355
 356             fscal            = felec;
 357
 358             fscal            = _mm_and_pd(fscal,cutoff_mask);
 359
 360             /* Update vectorial force */
 361             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 362             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 363             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 364
 365             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 366             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 367             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 368
 369             }
 370
 371             /**************************
 372              * CALCULATE INTERACTIONS *
 373              **************************/
 374
 375             if (gmx_mm_any_lt(rsq30,rcutoff2))
 376             {
 377
 378             /* Compute parameters for interactions between i and j atoms */
 379             qq30             = _mm_mul_pd(iq3,jq0);
 380
 381             /* REACTION-FIELD ELECTROSTATICS */
 382             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 383             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 384
 385             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velec            = _mm_and_pd(velec,cutoff_mask);
 389             velecsum         = _mm_add_pd(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             fscal            = _mm_and_pd(fscal,cutoff_mask);
 394
 395             /* Update vectorial force */
 396             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 397             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 398             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 399
 400             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 401             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 402             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 403
 404             }
 405
 406             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 407
 408             /* Inner loop uses 182 flops */
 409         }
 410
 411         if(jidx<j_index_end)
 412         {
 413
 414             jnrA             = jjnr[jidx];
 415             j_coord_offsetA  = DIM*jnrA;
 416
 417             /* load j atom coordinates */
 418             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 419                                               &jx0,&jy0,&jz0);
 420
 421             /* Calculate displacement vector */
 422             dx00             = _mm_sub_pd(ix0,jx0);
 423             dy00             = _mm_sub_pd(iy0,jy0);
 424             dz00             = _mm_sub_pd(iz0,jz0);
 425             dx10             = _mm_sub_pd(ix1,jx0);
 426             dy10             = _mm_sub_pd(iy1,jy0);
 427             dz10             = _mm_sub_pd(iz1,jz0);
 428             dx20             = _mm_sub_pd(ix2,jx0);
 429             dy20             = _mm_sub_pd(iy2,jy0);
 430             dz20             = _mm_sub_pd(iz2,jz0);
 431             dx30             = _mm_sub_pd(ix3,jx0);
 432             dy30             = _mm_sub_pd(iy3,jy0);
 433             dz30             = _mm_sub_pd(iz3,jz0);
 434
 435             /* Calculate squared distance and things based on it */
 436             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 437             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 438             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 439             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 440
 441             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 442             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 443             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 444             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 445
 446             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 447             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 448             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 449             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 450
 451             /* Load parameters for j particles */
 452             jq0              = _mm_load_sd(charge+jnrA+0);
 453             vdwjidx0A        = 2*vdwtype[jnrA+0];
 454
 455             fjx0             = _mm_setzero_pd();
 456             fjy0             = _mm_setzero_pd();
 457             fjz0             = _mm_setzero_pd();
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             if (gmx_mm_any_lt(rsq00,rcutoff2))
 464             {
 465
 466             r00              = _mm_mul_pd(rsq00,rinv00);
 467
 468             /* Compute parameters for interactions between i and j atoms */
 469             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 470
 471             /* LENNARD-JONES DISPERSION/REPULSION */
 472
 473             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 474             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 475             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 476             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 477             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 478
 479             d                = _mm_sub_pd(r00,rswitch);
 480             d                = _mm_max_pd(d,_mm_setzero_pd());
 481             d2               = _mm_mul_pd(d,d);
 482             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 483
 484             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 485
 486             /* Evaluate switch function */
 487             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 488             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 489             vvdw             = _mm_mul_pd(vvdw,sw);
 490             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 494             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 495             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 496
 497             fscal            = fvdw;
 498
 499             fscal            = _mm_and_pd(fscal,cutoff_mask);
 500
 501             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 502
 503             /* Update vectorial force */
 504             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 505             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 506             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 507
 508             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 509             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 510             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 511
 512             }
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             if (gmx_mm_any_lt(rsq10,rcutoff2))
 519             {
 520
 521             /* Compute parameters for interactions between i and j atoms */
 522             qq10             = _mm_mul_pd(iq1,jq0);
 523
 524             /* REACTION-FIELD ELECTROSTATICS */
 525             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 526             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 527
 528             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm_and_pd(velec,cutoff_mask);
 532             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 533             velecsum         = _mm_add_pd(velecsum,velec);
 534
 535             fscal            = felec;
 536
 537             fscal            = _mm_and_pd(fscal,cutoff_mask);
 538
 539             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 540
 541             /* Update vectorial force */
 542             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 543             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 544             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 545
 546             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 547             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 548             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 549
 550             }
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             if (gmx_mm_any_lt(rsq20,rcutoff2))
 557             {
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             qq20             = _mm_mul_pd(iq2,jq0);
 561
 562             /* REACTION-FIELD ELECTROSTATICS */
 563             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 564             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 565
 566             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velec            = _mm_and_pd(velec,cutoff_mask);
 570             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 571             velecsum         = _mm_add_pd(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm_and_pd(fscal,cutoff_mask);
 576
 577             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 578
 579             /* Update vectorial force */
 580             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 581             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 582             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 583
 584             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 585             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 586             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 587
 588             }
 589
 590             /**************************
 591              * CALCULATE INTERACTIONS *
 592              **************************/
 593
 594             if (gmx_mm_any_lt(rsq30,rcutoff2))
 595             {
 596
 597             /* Compute parameters for interactions between i and j atoms */
 598             qq30             = _mm_mul_pd(iq3,jq0);
 599
 600             /* REACTION-FIELD ELECTROSTATICS */
 601             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 602             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 603
 604             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 605
 606             /* Update potential sum for this i atom from the interaction with this j atom. */
 607             velec            = _mm_and_pd(velec,cutoff_mask);
 608             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 609             velecsum         = _mm_add_pd(velecsum,velec);
 610
 611             fscal            = felec;
 612
 613             fscal            = _mm_and_pd(fscal,cutoff_mask);
 614
 615             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 616
 617             /* Update vectorial force */
 618             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 619             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 620             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 621
 622             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 623             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 624             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 625
 626             }
 627
 628             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 629
 630             /* Inner loop uses 182 flops */
 631         }
 632
 633         /* End of innermost loop */
 634
 635         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 636                                               f+i_coord_offset,fshift+i_shift_offset);
 637
 638         ggid                        = gid[iidx];
 639         /* Update potential energies */
 640         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 641         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 642
 643         /* Increment number of inner iterations */
 644         inneriter                  += j_index_end - j_index_start;
 645
 646         /* Outer loop uses 26 flops */
 647     }
 648
 649     /* Increment number of outer iterations */
 650     outeriter        += nri;
 651
 652     /* Update outer/inner flops */
 653
 654     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*182);
 655 }
 656 /*
 657  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_128_fma_double
 658  * Electrostatics interaction: ReactionField
 659  * VdW interaction:            LennardJones
 660  * Geometry:                   Water4-Particle
 661  * Calculate force/pot:        Force
 662  */
 663 void
 664 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_128_fma_double
 665                     (t_nblist                    * gmx_restrict       nlist,
 666                      rvec                        * gmx_restrict          xx,
 667                      rvec                        * gmx_restrict          ff,
 668                      t_forcerec                  * gmx_restrict          fr,
 669                      t_mdatoms                   * gmx_restrict     mdatoms,
 670                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 671                      t_nrnb                      * gmx_restrict        nrnb)
 672 {
 673     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 674      * just 0 for non-waters.
 675      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 676      * jnr indices corresponding to data put in the four positions in the SIMD register.
 677      */
 678     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 679     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 680     int              jnrA,jnrB;
 681     int              j_coord_offsetA,j_coord_offsetB;
 682     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 683     real             rcutoff_scalar;
 684     real             *shiftvec,*fshift,*x,*f;
 685     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 686     int              vdwioffset0;
 687     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 688     int              vdwioffset1;
 689     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 690     int              vdwioffset2;
 691     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 692     int              vdwioffset3;
 693     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 694     int              vdwjidx0A,vdwjidx0B;
 695     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 696     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 697     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 698     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 699     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 700     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 701     real             *charge;
 702     int              nvdwtype;
 703     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 704     int              *vdwtype;
 705     real             *vdwparam;
 706     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 707     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 708     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 709     real             rswitch_scalar,d_scalar;
 710     __m128d          dummy_mask,cutoff_mask;
 711     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 712     __m128d          one     = _mm_set1_pd(1.0);
 713     __m128d          two     = _mm_set1_pd(2.0);
 714     x                = xx[0];
 715     f                = ff[0];
 716
 717     nri              = nlist->nri;
 718     iinr             = nlist->iinr;
 719     jindex           = nlist->jindex;
 720     jjnr             = nlist->jjnr;
 721     shiftidx         = nlist->shift;
 722     gid              = nlist->gid;
 723     shiftvec         = fr->shift_vec[0];
 724     fshift           = fr->fshift[0];
 725     facel            = _mm_set1_pd(fr->epsfac);
 726     charge           = mdatoms->chargeA;
 727     krf              = _mm_set1_pd(fr->ic->k_rf);
 728     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 729     crf              = _mm_set1_pd(fr->ic->c_rf);
 730     nvdwtype         = fr->ntype;
 731     vdwparam         = fr->nbfp;
 732     vdwtype          = mdatoms->typeA;
 733
 734     /* Setup water-specific parameters */
 735     inr              = nlist->iinr[0];
 736     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 737     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 738     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 739     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 740
 741     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 742     rcutoff_scalar   = fr->rcoulomb;
 743     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 744     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 745
 746     rswitch_scalar   = fr->rvdw_switch;
 747     rswitch          = _mm_set1_pd(rswitch_scalar);
 748     /* Setup switch parameters */
 749     d_scalar         = rcutoff_scalar-rswitch_scalar;
 750     d                = _mm_set1_pd(d_scalar);
 751     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 752     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 753     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 754     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 755     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 756     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 757
 758     /* Avoid stupid compiler warnings */
 759     jnrA = jnrB = 0;
 760     j_coord_offsetA = 0;
 761     j_coord_offsetB = 0;
 762
 763     outeriter        = 0;
 764     inneriter        = 0;
 765
 766     /* Start outer loop over neighborlists */
 767     for(iidx=0; iidx<nri; iidx++)
 768     {
 769         /* Load shift vector for this list */
 770         i_shift_offset   = DIM*shiftidx[iidx];
 771
 772         /* Load limits for loop over neighbors */
 773         j_index_start    = jindex[iidx];
 774         j_index_end      = jindex[iidx+1];
 775
 776         /* Get outer coordinate index */
 777         inr              = iinr[iidx];
 778         i_coord_offset   = DIM*inr;
 779
 780         /* Load i particle coords and add shift vector */
 781         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 782                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 783
 784         fix0             = _mm_setzero_pd();
 785         fiy0             = _mm_setzero_pd();
 786         fiz0             = _mm_setzero_pd();
 787         fix1             = _mm_setzero_pd();
 788         fiy1             = _mm_setzero_pd();
 789         fiz1             = _mm_setzero_pd();
 790         fix2             = _mm_setzero_pd();
 791         fiy2             = _mm_setzero_pd();
 792         fiz2             = _mm_setzero_pd();
 793         fix3             = _mm_setzero_pd();
 794         fiy3             = _mm_setzero_pd();
 795         fiz3             = _mm_setzero_pd();
 796
 797         /* Start inner kernel loop */
 798         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 799         {
 800
 801             /* Get j neighbor index, and coordinate index */
 802             jnrA             = jjnr[jidx];
 803             jnrB             = jjnr[jidx+1];
 804             j_coord_offsetA  = DIM*jnrA;
 805             j_coord_offsetB  = DIM*jnrB;
 806
 807             /* load j atom coordinates */
 808             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 809                                               &jx0,&jy0,&jz0);
 810
 811             /* Calculate displacement vector */
 812             dx00             = _mm_sub_pd(ix0,jx0);
 813             dy00             = _mm_sub_pd(iy0,jy0);
 814             dz00             = _mm_sub_pd(iz0,jz0);
 815             dx10             = _mm_sub_pd(ix1,jx0);
 816             dy10             = _mm_sub_pd(iy1,jy0);
 817             dz10             = _mm_sub_pd(iz1,jz0);
 818             dx20             = _mm_sub_pd(ix2,jx0);
 819             dy20             = _mm_sub_pd(iy2,jy0);
 820             dz20             = _mm_sub_pd(iz2,jz0);
 821             dx30             = _mm_sub_pd(ix3,jx0);
 822             dy30             = _mm_sub_pd(iy3,jy0);
 823             dz30             = _mm_sub_pd(iz3,jz0);
 824
 825             /* Calculate squared distance and things based on it */
 826             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 827             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 828             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 829             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 830
 831             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 832             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 833             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 834             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 835
 836             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 837             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 838             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 839             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 840
 841             /* Load parameters for j particles */
 842             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 843             vdwjidx0A        = 2*vdwtype[jnrA+0];
 844             vdwjidx0B        = 2*vdwtype[jnrB+0];
 845
 846             fjx0             = _mm_setzero_pd();
 847             fjy0             = _mm_setzero_pd();
 848             fjz0             = _mm_setzero_pd();
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             if (gmx_mm_any_lt(rsq00,rcutoff2))
 855             {
 856
 857             r00              = _mm_mul_pd(rsq00,rinv00);
 858
 859             /* Compute parameters for interactions between i and j atoms */
 860             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 861                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 862
 863             /* LENNARD-JONES DISPERSION/REPULSION */
 864
 865             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 866             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 867             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 868             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 869             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 870
 871             d                = _mm_sub_pd(r00,rswitch);
 872             d                = _mm_max_pd(d,_mm_setzero_pd());
 873             d2               = _mm_mul_pd(d,d);
 874             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 875
 876             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 877
 878             /* Evaluate switch function */
 879             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 880             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 881             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 882
 883             fscal            = fvdw;
 884
 885             fscal            = _mm_and_pd(fscal,cutoff_mask);
 886
 887             /* Update vectorial force */
 888             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 889             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 890             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 891
 892             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 893             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 894             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 895
 896             }
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             if (gmx_mm_any_lt(rsq10,rcutoff2))
 903             {
 904
 905             /* Compute parameters for interactions between i and j atoms */
 906             qq10             = _mm_mul_pd(iq1,jq0);
 907
 908             /* REACTION-FIELD ELECTROSTATICS */
 909             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 910
 911             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 912
 913             fscal            = felec;
 914
 915             fscal            = _mm_and_pd(fscal,cutoff_mask);
 916
 917             /* Update vectorial force */
 918             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 919             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 920             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 921
 922             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 923             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 924             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 925
 926             }
 927
 928             /**************************
 929              * CALCULATE INTERACTIONS *
 930              **************************/
 931
 932             if (gmx_mm_any_lt(rsq20,rcutoff2))
 933             {
 934
 935             /* Compute parameters for interactions between i and j atoms */
 936             qq20             = _mm_mul_pd(iq2,jq0);
 937
 938             /* REACTION-FIELD ELECTROSTATICS */
 939             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 940
 941             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 942
 943             fscal            = felec;
 944
 945             fscal            = _mm_and_pd(fscal,cutoff_mask);
 946
 947             /* Update vectorial force */
 948             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 949             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 950             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 951
 952             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 953             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 954             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 955
 956             }
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             if (gmx_mm_any_lt(rsq30,rcutoff2))
 963             {
 964
 965             /* Compute parameters for interactions between i and j atoms */
 966             qq30             = _mm_mul_pd(iq3,jq0);
 967
 968             /* REACTION-FIELD ELECTROSTATICS */
 969             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 970
 971             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm_and_pd(fscal,cutoff_mask);
 976
 977             /* Update vectorial force */
 978             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 979             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 980             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 981
 982             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 983             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 984             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 985
 986             }
 987
 988             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 989
 990             /* Inner loop uses 161 flops */
 991         }
 992
 993         if(jidx<j_index_end)
 994         {
 995
 996             jnrA             = jjnr[jidx];
 997             j_coord_offsetA  = DIM*jnrA;
 998
 999             /* load j atom coordinates */
1000             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1001                                               &jx0,&jy0,&jz0);
1002
1003             /* Calculate displacement vector */
1004             dx00             = _mm_sub_pd(ix0,jx0);
1005             dy00             = _mm_sub_pd(iy0,jy0);
1006             dz00             = _mm_sub_pd(iz0,jz0);
1007             dx10             = _mm_sub_pd(ix1,jx0);
1008             dy10             = _mm_sub_pd(iy1,jy0);
1009             dz10             = _mm_sub_pd(iz1,jz0);
1010             dx20             = _mm_sub_pd(ix2,jx0);
1011             dy20             = _mm_sub_pd(iy2,jy0);
1012             dz20             = _mm_sub_pd(iz2,jz0);
1013             dx30             = _mm_sub_pd(ix3,jx0);
1014             dy30             = _mm_sub_pd(iy3,jy0);
1015             dz30             = _mm_sub_pd(iz3,jz0);
1016
1017             /* Calculate squared distance and things based on it */
1018             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1019             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1020             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1021             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1022
1023             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1024             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1025             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1026             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1027
1028             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1029             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1030             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1031             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1032
1033             /* Load parameters for j particles */
1034             jq0              = _mm_load_sd(charge+jnrA+0);
1035             vdwjidx0A        = 2*vdwtype[jnrA+0];
1036
1037             fjx0             = _mm_setzero_pd();
1038             fjy0             = _mm_setzero_pd();
1039             fjz0             = _mm_setzero_pd();
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             if (gmx_mm_any_lt(rsq00,rcutoff2))
1046             {
1047
1048             r00              = _mm_mul_pd(rsq00,rinv00);
1049
1050             /* Compute parameters for interactions between i and j atoms */
1051             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1052
1053             /* LENNARD-JONES DISPERSION/REPULSION */
1054
1055             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1056             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1057             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1058             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
1059             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1060
1061             d                = _mm_sub_pd(r00,rswitch);
1062             d                = _mm_max_pd(d,_mm_setzero_pd());
1063             d2               = _mm_mul_pd(d,d);
1064             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1065
1066             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1067
1068             /* Evaluate switch function */
1069             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1070             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1071             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1072
1073             fscal            = fvdw;
1074
1075             fscal            = _mm_and_pd(fscal,cutoff_mask);
1076
1077             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1078
1079             /* Update vectorial force */
1080             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1081             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1082             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1083
1084             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1085             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1086             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1087
1088             }
1089
1090             /**************************
1091              * CALCULATE INTERACTIONS *
1092              **************************/
1093
1094             if (gmx_mm_any_lt(rsq10,rcutoff2))
1095             {
1096
1097             /* Compute parameters for interactions between i and j atoms */
1098             qq10             = _mm_mul_pd(iq1,jq0);
1099
1100             /* REACTION-FIELD ELECTROSTATICS */
1101             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
1102
1103             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm_and_pd(fscal,cutoff_mask);
1108
1109             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1110
1111             /* Update vectorial force */
1112             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1113             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1114             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1115
1116             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1117             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1118             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1119
1120             }
1121
1122             /**************************
1123              * CALCULATE INTERACTIONS *
1124              **************************/
1125
1126             if (gmx_mm_any_lt(rsq20,rcutoff2))
1127             {
1128
1129             /* Compute parameters for interactions between i and j atoms */
1130             qq20             = _mm_mul_pd(iq2,jq0);
1131
1132             /* REACTION-FIELD ELECTROSTATICS */
1133             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1134
1135             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1136
1137             fscal            = felec;
1138
1139             fscal            = _mm_and_pd(fscal,cutoff_mask);
1140
1141             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1142
1143             /* Update vectorial force */
1144             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1145             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1146             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1147
1148             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1149             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1150             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1151
1152             }
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             if (gmx_mm_any_lt(rsq30,rcutoff2))
1159             {
1160
1161             /* Compute parameters for interactions between i and j atoms */
1162             qq30             = _mm_mul_pd(iq3,jq0);
1163
1164             /* REACTION-FIELD ELECTROSTATICS */
1165             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
1166
1167             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1168
1169             fscal            = felec;
1170
1171             fscal            = _mm_and_pd(fscal,cutoff_mask);
1172
1173             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1174
1175             /* Update vectorial force */
1176             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1177             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1178             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1179
1180             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1181             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1182             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1183
1184             }
1185
1186             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1187
1188             /* Inner loop uses 161 flops */
1189         }
1190
1191         /* End of innermost loop */
1192
1193         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1194                                               f+i_coord_offset,fshift+i_shift_offset);
1195
1196         /* Increment number of inner iterations */
1197         inneriter                  += j_index_end - j_index_start;
1198
1199         /* Outer loop uses 24 flops */
1200     }
1201
1202     /* Increment number of outer iterations */
1203     outeriter        += nri;
1204
1205     /* Update outer/inner flops */
1206
1207     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*161);
1208 }