src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(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_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     krf              = _mm_set1_ps(fr->ic->k_rf);
 122     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 123     crf              = _mm_set1_ps(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     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 131     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 132     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 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_ps(rcutoff_scalar);
 138     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 139
 140     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 141     rvdw             = _mm_set1_ps(fr->rvdw);
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 175
 176         fix0             = _mm_setzero_ps();
 177         fiy0             = _mm_setzero_ps();
 178         fiz0             = _mm_setzero_ps();
 179         fix1             = _mm_setzero_ps();
 180         fiy1             = _mm_setzero_ps();
 181         fiz1             = _mm_setzero_ps();
 182         fix2             = _mm_setzero_ps();
 183         fiy2             = _mm_setzero_ps();
 184         fiz2             = _mm_setzero_ps();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_ps();
 188         vvdwsum          = _mm_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               x+j_coord_offsetC,x+j_coord_offsetD,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_ps(ix0,jx0);
 211             dy00             = _mm_sub_ps(iy0,jy0);
 212             dz00             = _mm_sub_ps(iz0,jz0);
 213             dx10             = _mm_sub_ps(ix1,jx0);
 214             dy10             = _mm_sub_ps(iy1,jy0);
 215             dz10             = _mm_sub_ps(iz1,jz0);
 216             dx20             = _mm_sub_ps(ix2,jx0);
 217             dy20             = _mm_sub_ps(iy2,jy0);
 218             dz20             = _mm_sub_ps(iz2,jz0);
 219
 220             /* Calculate squared distance and things based on it */
 221             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 222             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 223             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 224
 225             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 226             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 227             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 228
 229             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 230             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 231             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 235                                                               charge+jnrC+0,charge+jnrD+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240
 241             fjx0             = _mm_setzero_ps();
 242             fjy0             = _mm_setzero_ps();
 243             fjz0             = _mm_setzero_ps();
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             if (gmx_mm_any_lt(rsq00,rcutoff2))
 250             {
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             qq00             = _mm_mul_ps(iq0,jq0);
 254             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 255                                          vdwparam+vdwioffset0+vdwjidx0B,
 256                                          vdwparam+vdwioffset0+vdwjidx0C,
 257                                          vdwparam+vdwioffset0+vdwjidx0D,
 258                                          &c6_00,&c12_00);
 259
 260             /* REACTION-FIELD ELECTROSTATICS */
 261             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 262             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 263
 264             /* LENNARD-JONES DISPERSION/REPULSION */
 265
 266             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 267             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 268             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 269             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 270                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 271             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 272
 273             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velec            = _mm_and_ps(velec,cutoff_mask);
 277             velecsum         = _mm_add_ps(velecsum,velec);
 278             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 279             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 280
 281             fscal            = _mm_add_ps(felec,fvdw);
 282
 283             fscal            = _mm_and_ps(fscal,cutoff_mask);
 284
 285              /* Update vectorial force */
 286             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 287             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 288             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 289
 290             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 291             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 292             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 293
 294             }
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             if (gmx_mm_any_lt(rsq10,rcutoff2))
 301             {
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm_mul_ps(iq1,jq0);
 305
 306             /* REACTION-FIELD ELECTROSTATICS */
 307             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 308             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 309
 310             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velec            = _mm_and_ps(velec,cutoff_mask);
 314             velecsum         = _mm_add_ps(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             fscal            = _mm_and_ps(fscal,cutoff_mask);
 319
 320              /* Update vectorial force */
 321             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 322             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 323             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 324
 325             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 326             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 327             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 328
 329             }
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             if (gmx_mm_any_lt(rsq20,rcutoff2))
 336             {
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq20             = _mm_mul_ps(iq2,jq0);
 340
 341             /* REACTION-FIELD ELECTROSTATICS */
 342             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 343             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 344
 345             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velec            = _mm_and_ps(velec,cutoff_mask);
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             fscal            = _mm_and_ps(fscal,cutoff_mask);
 354
 355              /* Update vectorial force */
 356             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 357             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 358             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 359
 360             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 361             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 362             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 363
 364             }
 365
 366             fjptrA             = f+j_coord_offsetA;
 367             fjptrB             = f+j_coord_offsetB;
 368             fjptrC             = f+j_coord_offsetC;
 369             fjptrD             = f+j_coord_offsetD;
 370
 371             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 372
 373             /* Inner loop uses 135 flops */
 374         }
 375
 376         if(jidx<j_index_end)
 377         {
 378
 379             /* Get j neighbor index, and coordinate index */
 380             jnrlistA         = jjnr[jidx];
 381             jnrlistB         = jjnr[jidx+1];
 382             jnrlistC         = jjnr[jidx+2];
 383             jnrlistD         = jjnr[jidx+3];
 384             /* Sign of each element will be negative for non-real atoms.
 385              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 386              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 387              */
 388             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 389             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 390             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 391             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 392             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 393             j_coord_offsetA  = DIM*jnrA;
 394             j_coord_offsetB  = DIM*jnrB;
 395             j_coord_offsetC  = DIM*jnrC;
 396             j_coord_offsetD  = DIM*jnrD;
 397
 398             /* load j atom coordinates */
 399             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 400                                               x+j_coord_offsetC,x+j_coord_offsetD,
 401                                               &jx0,&jy0,&jz0);
 402
 403             /* Calculate displacement vector */
 404             dx00             = _mm_sub_ps(ix0,jx0);
 405             dy00             = _mm_sub_ps(iy0,jy0);
 406             dz00             = _mm_sub_ps(iz0,jz0);
 407             dx10             = _mm_sub_ps(ix1,jx0);
 408             dy10             = _mm_sub_ps(iy1,jy0);
 409             dz10             = _mm_sub_ps(iz1,jz0);
 410             dx20             = _mm_sub_ps(ix2,jx0);
 411             dy20             = _mm_sub_ps(iy2,jy0);
 412             dz20             = _mm_sub_ps(iz2,jz0);
 413
 414             /* Calculate squared distance and things based on it */
 415             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 416             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 417             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 418
 419             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 420             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 421             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 422
 423             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 424             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 425             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 426
 427             /* Load parameters for j particles */
 428             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 429                                                               charge+jnrC+0,charge+jnrD+0);
 430             vdwjidx0A        = 2*vdwtype[jnrA+0];
 431             vdwjidx0B        = 2*vdwtype[jnrB+0];
 432             vdwjidx0C        = 2*vdwtype[jnrC+0];
 433             vdwjidx0D        = 2*vdwtype[jnrD+0];
 434
 435             fjx0             = _mm_setzero_ps();
 436             fjy0             = _mm_setzero_ps();
 437             fjz0             = _mm_setzero_ps();
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             if (gmx_mm_any_lt(rsq00,rcutoff2))
 444             {
 445
 446             /* Compute parameters for interactions between i and j atoms */
 447             qq00             = _mm_mul_ps(iq0,jq0);
 448             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 449                                          vdwparam+vdwioffset0+vdwjidx0B,
 450                                          vdwparam+vdwioffset0+vdwjidx0C,
 451                                          vdwparam+vdwioffset0+vdwjidx0D,
 452                                          &c6_00,&c12_00);
 453
 454             /* REACTION-FIELD ELECTROSTATICS */
 455             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 456             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 457
 458             /* LENNARD-JONES DISPERSION/REPULSION */
 459
 460             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 461             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 462             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 463             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 464                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 465             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 466
 467             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 468
 469             /* Update potential sum for this i atom from the interaction with this j atom. */
 470             velec            = _mm_and_ps(velec,cutoff_mask);
 471             velec            = _mm_andnot_ps(dummy_mask,velec);
 472             velecsum         = _mm_add_ps(velecsum,velec);
 473             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 474             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 475             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 476
 477             fscal            = _mm_add_ps(felec,fvdw);
 478
 479             fscal            = _mm_and_ps(fscal,cutoff_mask);
 480
 481             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 482
 483              /* Update vectorial force */
 484             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 485             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 486             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 487
 488             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 489             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 490             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 491
 492             }
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             if (gmx_mm_any_lt(rsq10,rcutoff2))
 499             {
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             qq10             = _mm_mul_ps(iq1,jq0);
 503
 504             /* REACTION-FIELD ELECTROSTATICS */
 505             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 506             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 507
 508             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             velec            = _mm_and_ps(velec,cutoff_mask);
 512             velec            = _mm_andnot_ps(dummy_mask,velec);
 513             velecsum         = _mm_add_ps(velecsum,velec);
 514
 515             fscal            = felec;
 516
 517             fscal            = _mm_and_ps(fscal,cutoff_mask);
 518
 519             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 520
 521              /* Update vectorial force */
 522             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 523             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 524             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 525
 526             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 527             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 528             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 529
 530             }
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             if (gmx_mm_any_lt(rsq20,rcutoff2))
 537             {
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq20             = _mm_mul_ps(iq2,jq0);
 541
 542             /* REACTION-FIELD ELECTROSTATICS */
 543             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 544             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 545
 546             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             velec            = _mm_and_ps(velec,cutoff_mask);
 550             velec            = _mm_andnot_ps(dummy_mask,velec);
 551             velecsum         = _mm_add_ps(velecsum,velec);
 552
 553             fscal            = felec;
 554
 555             fscal            = _mm_and_ps(fscal,cutoff_mask);
 556
 557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 558
 559              /* Update vectorial force */
 560             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 561             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 562             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 563
 564             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 565             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 566             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 567
 568             }
 569
 570             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 571             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 572             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 573             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 574
 575             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 576
 577             /* Inner loop uses 135 flops */
 578         }
 579
 580         /* End of innermost loop */
 581
 582         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 583                                               f+i_coord_offset,fshift+i_shift_offset);
 584
 585         ggid                        = gid[iidx];
 586         /* Update potential energies */
 587         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 588         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 589
 590         /* Increment number of inner iterations */
 591         inneriter                  += j_index_end - j_index_start;
 592
 593         /* Outer loop uses 20 flops */
 594     }
 595
 596     /* Increment number of outer iterations */
 597     outeriter        += nri;
 598
 599     /* Update outer/inner flops */
 600
 601     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*135);
 602 }
 603 /*
 604  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_128_fma_single
 605  * Electrostatics interaction: ReactionField
 606  * VdW interaction:            LennardJones
 607  * Geometry:                   Water3-Particle
 608  * Calculate force/pot:        Force
 609  */
 610 void
 611 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_128_fma_single
 612                     (t_nblist                    * gmx_restrict       nlist,
 613                      rvec                        * gmx_restrict          xx,
 614                      rvec                        * gmx_restrict          ff,
 615                      t_forcerec                  * gmx_restrict          fr,
 616                      t_mdatoms                   * gmx_restrict     mdatoms,
 617                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 618                      t_nrnb                      * gmx_restrict        nrnb)
 619 {
 620     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 621      * just 0 for non-waters.
 622      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 623      * jnr indices corresponding to data put in the four positions in the SIMD register.
 624      */
 625     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 626     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 627     int              jnrA,jnrB,jnrC,jnrD;
 628     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 629     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 630     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 631     real             rcutoff_scalar;
 632     real             *shiftvec,*fshift,*x,*f;
 633     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 634     real             scratch[4*DIM];
 635     __m128           fscal,rcutoff,rcutoff2,jidxall;
 636     int              vdwioffset0;
 637     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 638     int              vdwioffset1;
 639     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 640     int              vdwioffset2;
 641     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 642     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 643     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 644     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 645     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 646     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 647     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 648     real             *charge;
 649     int              nvdwtype;
 650     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 651     int              *vdwtype;
 652     real             *vdwparam;
 653     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 654     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 655     __m128           dummy_mask,cutoff_mask;
 656     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 657     __m128           one     = _mm_set1_ps(1.0);
 658     __m128           two     = _mm_set1_ps(2.0);
 659     x                = xx[0];
 660     f                = ff[0];
 661
 662     nri              = nlist->nri;
 663     iinr             = nlist->iinr;
 664     jindex           = nlist->jindex;
 665     jjnr             = nlist->jjnr;
 666     shiftidx         = nlist->shift;
 667     gid              = nlist->gid;
 668     shiftvec         = fr->shift_vec[0];
 669     fshift           = fr->fshift[0];
 670     facel            = _mm_set1_ps(fr->epsfac);
 671     charge           = mdatoms->chargeA;
 672     krf              = _mm_set1_ps(fr->ic->k_rf);
 673     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 674     crf              = _mm_set1_ps(fr->ic->c_rf);
 675     nvdwtype         = fr->ntype;
 676     vdwparam         = fr->nbfp;
 677     vdwtype          = mdatoms->typeA;
 678
 679     /* Setup water-specific parameters */
 680     inr              = nlist->iinr[0];
 681     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 682     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 683     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 684     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 685
 686     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 687     rcutoff_scalar   = fr->rcoulomb;
 688     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 689     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 690
 691     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 692     rvdw             = _mm_set1_ps(fr->rvdw);
 693
 694     /* Avoid stupid compiler warnings */
 695     jnrA = jnrB = jnrC = jnrD = 0;
 696     j_coord_offsetA = 0;
 697     j_coord_offsetB = 0;
 698     j_coord_offsetC = 0;
 699     j_coord_offsetD = 0;
 700
 701     outeriter        = 0;
 702     inneriter        = 0;
 703
 704     for(iidx=0;iidx<4*DIM;iidx++)
 705     {
 706         scratch[iidx] = 0.0;
 707     }
 708
 709     /* Start outer loop over neighborlists */
 710     for(iidx=0; iidx<nri; iidx++)
 711     {
 712         /* Load shift vector for this list */
 713         i_shift_offset   = DIM*shiftidx[iidx];
 714
 715         /* Load limits for loop over neighbors */
 716         j_index_start    = jindex[iidx];
 717         j_index_end      = jindex[iidx+1];
 718
 719         /* Get outer coordinate index */
 720         inr              = iinr[iidx];
 721         i_coord_offset   = DIM*inr;
 722
 723         /* Load i particle coords and add shift vector */
 724         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 725                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 726
 727         fix0             = _mm_setzero_ps();
 728         fiy0             = _mm_setzero_ps();
 729         fiz0             = _mm_setzero_ps();
 730         fix1             = _mm_setzero_ps();
 731         fiy1             = _mm_setzero_ps();
 732         fiz1             = _mm_setzero_ps();
 733         fix2             = _mm_setzero_ps();
 734         fiy2             = _mm_setzero_ps();
 735         fiz2             = _mm_setzero_ps();
 736
 737         /* Start inner kernel loop */
 738         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 739         {
 740
 741             /* Get j neighbor index, and coordinate index */
 742             jnrA             = jjnr[jidx];
 743             jnrB             = jjnr[jidx+1];
 744             jnrC             = jjnr[jidx+2];
 745             jnrD             = jjnr[jidx+3];
 746             j_coord_offsetA  = DIM*jnrA;
 747             j_coord_offsetB  = DIM*jnrB;
 748             j_coord_offsetC  = DIM*jnrC;
 749             j_coord_offsetD  = DIM*jnrD;
 750
 751             /* load j atom coordinates */
 752             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 753                                               x+j_coord_offsetC,x+j_coord_offsetD,
 754                                               &jx0,&jy0,&jz0);
 755
 756             /* Calculate displacement vector */
 757             dx00             = _mm_sub_ps(ix0,jx0);
 758             dy00             = _mm_sub_ps(iy0,jy0);
 759             dz00             = _mm_sub_ps(iz0,jz0);
 760             dx10             = _mm_sub_ps(ix1,jx0);
 761             dy10             = _mm_sub_ps(iy1,jy0);
 762             dz10             = _mm_sub_ps(iz1,jz0);
 763             dx20             = _mm_sub_ps(ix2,jx0);
 764             dy20             = _mm_sub_ps(iy2,jy0);
 765             dz20             = _mm_sub_ps(iz2,jz0);
 766
 767             /* Calculate squared distance and things based on it */
 768             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 769             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 770             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 771
 772             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 773             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 774             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 775
 776             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 777             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 778             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 779
 780             /* Load parameters for j particles */
 781             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 782                                                               charge+jnrC+0,charge+jnrD+0);
 783             vdwjidx0A        = 2*vdwtype[jnrA+0];
 784             vdwjidx0B        = 2*vdwtype[jnrB+0];
 785             vdwjidx0C        = 2*vdwtype[jnrC+0];
 786             vdwjidx0D        = 2*vdwtype[jnrD+0];
 787
 788             fjx0             = _mm_setzero_ps();
 789             fjy0             = _mm_setzero_ps();
 790             fjz0             = _mm_setzero_ps();
 791
 792             /**************************
 793              * CALCULATE INTERACTIONS *
 794              **************************/
 795
 796             if (gmx_mm_any_lt(rsq00,rcutoff2))
 797             {
 798
 799             /* Compute parameters for interactions between i and j atoms */
 800             qq00             = _mm_mul_ps(iq0,jq0);
 801             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 802                                          vdwparam+vdwioffset0+vdwjidx0B,
 803                                          vdwparam+vdwioffset0+vdwjidx0C,
 804                                          vdwparam+vdwioffset0+vdwjidx0D,
 805                                          &c6_00,&c12_00);
 806
 807             /* REACTION-FIELD ELECTROSTATICS */
 808             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 809
 810             /* LENNARD-JONES DISPERSION/REPULSION */
 811
 812             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 813             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 814
 815             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 816
 817             fscal            = _mm_add_ps(felec,fvdw);
 818
 819             fscal            = _mm_and_ps(fscal,cutoff_mask);
 820
 821              /* Update vectorial force */
 822             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 823             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 824             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 825
 826             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 827             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 828             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 829
 830             }
 831
 832             /**************************
 833              * CALCULATE INTERACTIONS *
 834              **************************/
 835
 836             if (gmx_mm_any_lt(rsq10,rcutoff2))
 837             {
 838
 839             /* Compute parameters for interactions between i and j atoms */
 840             qq10             = _mm_mul_ps(iq1,jq0);
 841
 842             /* REACTION-FIELD ELECTROSTATICS */
 843             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 844
 845             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 846
 847             fscal            = felec;
 848
 849             fscal            = _mm_and_ps(fscal,cutoff_mask);
 850
 851              /* Update vectorial force */
 852             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 853             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 854             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 855
 856             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 857             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 858             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 859
 860             }
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             if (gmx_mm_any_lt(rsq20,rcutoff2))
 867             {
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq20             = _mm_mul_ps(iq2,jq0);
 871
 872             /* REACTION-FIELD ELECTROSTATICS */
 873             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 874
 875             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 876
 877             fscal            = felec;
 878
 879             fscal            = _mm_and_ps(fscal,cutoff_mask);
 880
 881              /* Update vectorial force */
 882             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 883             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 884             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 885
 886             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 887             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 888             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 889
 890             }
 891
 892             fjptrA             = f+j_coord_offsetA;
 893             fjptrB             = f+j_coord_offsetB;
 894             fjptrC             = f+j_coord_offsetC;
 895             fjptrD             = f+j_coord_offsetD;
 896
 897             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 898
 899             /* Inner loop uses 106 flops */
 900         }
 901
 902         if(jidx<j_index_end)
 903         {
 904
 905             /* Get j neighbor index, and coordinate index */
 906             jnrlistA         = jjnr[jidx];
 907             jnrlistB         = jjnr[jidx+1];
 908             jnrlistC         = jjnr[jidx+2];
 909             jnrlistD         = jjnr[jidx+3];
 910             /* Sign of each element will be negative for non-real atoms.
 911              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 912              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 913              */
 914             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 915             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 916             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 917             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 918             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 919             j_coord_offsetA  = DIM*jnrA;
 920             j_coord_offsetB  = DIM*jnrB;
 921             j_coord_offsetC  = DIM*jnrC;
 922             j_coord_offsetD  = DIM*jnrD;
 923
 924             /* load j atom coordinates */
 925             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 926                                               x+j_coord_offsetC,x+j_coord_offsetD,
 927                                               &jx0,&jy0,&jz0);
 928
 929             /* Calculate displacement vector */
 930             dx00             = _mm_sub_ps(ix0,jx0);
 931             dy00             = _mm_sub_ps(iy0,jy0);
 932             dz00             = _mm_sub_ps(iz0,jz0);
 933             dx10             = _mm_sub_ps(ix1,jx0);
 934             dy10             = _mm_sub_ps(iy1,jy0);
 935             dz10             = _mm_sub_ps(iz1,jz0);
 936             dx20             = _mm_sub_ps(ix2,jx0);
 937             dy20             = _mm_sub_ps(iy2,jy0);
 938             dz20             = _mm_sub_ps(iz2,jz0);
 939
 940             /* Calculate squared distance and things based on it */
 941             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 942             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 943             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 944
 945             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 946             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 947             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 948
 949             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 950             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 951             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 952
 953             /* Load parameters for j particles */
 954             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 955                                                               charge+jnrC+0,charge+jnrD+0);
 956             vdwjidx0A        = 2*vdwtype[jnrA+0];
 957             vdwjidx0B        = 2*vdwtype[jnrB+0];
 958             vdwjidx0C        = 2*vdwtype[jnrC+0];
 959             vdwjidx0D        = 2*vdwtype[jnrD+0];
 960
 961             fjx0             = _mm_setzero_ps();
 962             fjy0             = _mm_setzero_ps();
 963             fjz0             = _mm_setzero_ps();
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             if (gmx_mm_any_lt(rsq00,rcutoff2))
 970             {
 971
 972             /* Compute parameters for interactions between i and j atoms */
 973             qq00             = _mm_mul_ps(iq0,jq0);
 974             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 975                                          vdwparam+vdwioffset0+vdwjidx0B,
 976                                          vdwparam+vdwioffset0+vdwjidx0C,
 977                                          vdwparam+vdwioffset0+vdwjidx0D,
 978                                          &c6_00,&c12_00);
 979
 980             /* REACTION-FIELD ELECTROSTATICS */
 981             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 982
 983             /* LENNARD-JONES DISPERSION/REPULSION */
 984
 985             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 986             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 987
 988             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 989
 990             fscal            = _mm_add_ps(felec,fvdw);
 991
 992             fscal            = _mm_and_ps(fscal,cutoff_mask);
 993
 994             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 995
 996              /* Update vectorial force */
 997             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 998             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 999             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1000
1001             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1002             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1003             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1004
1005             }
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             if (gmx_mm_any_lt(rsq10,rcutoff2))
1012             {
1013
1014             /* Compute parameters for interactions between i and j atoms */
1015             qq10             = _mm_mul_ps(iq1,jq0);
1016
1017             /* REACTION-FIELD ELECTROSTATICS */
1018             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1019
1020             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1021
1022             fscal            = felec;
1023
1024             fscal            = _mm_and_ps(fscal,cutoff_mask);
1025
1026             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1027
1028              /* Update vectorial force */
1029             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1030             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1031             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1032
1033             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1034             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1035             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1036
1037             }
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             if (gmx_mm_any_lt(rsq20,rcutoff2))
1044             {
1045
1046             /* Compute parameters for interactions between i and j atoms */
1047             qq20             = _mm_mul_ps(iq2,jq0);
1048
1049             /* REACTION-FIELD ELECTROSTATICS */
1050             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1051
1052             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1053
1054             fscal            = felec;
1055
1056             fscal            = _mm_and_ps(fscal,cutoff_mask);
1057
1058             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1059
1060              /* Update vectorial force */
1061             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1062             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1063             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1064
1065             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1066             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1067             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1068
1069             }
1070
1071             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1072             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1073             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1074             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1075
1076             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1077
1078             /* Inner loop uses 106 flops */
1079         }
1080
1081         /* End of innermost loop */
1082
1083         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1084                                               f+i_coord_offset,fshift+i_shift_offset);
1085
1086         /* Increment number of inner iterations */
1087         inneriter                  += j_index_end - j_index_start;
1088
1089         /* Outer loop uses 18 flops */
1090     }
1091
1092     /* Increment number of outer iterations */
1093     outeriter        += nri;
1094
1095     /* Update outer/inner flops */
1096
1097     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*106);
1098 }