src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwLJ_GeomW4P1_avx_128_fma_single.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_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecCoul_VdwLJ_GeomW4P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_avx_128_fma_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              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128           dummy_mask,cutoff_mask;
 108     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 109     __m128           one     = _mm_set1_ps(1.0);
 110     __m128           two     = _mm_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 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_ps(facel,_mm_set1_ps(charge[inr+1]));
 131     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 132     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 133     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 167
 168         fix0             = _mm_setzero_ps();
 169         fiy0             = _mm_setzero_ps();
 170         fiz0             = _mm_setzero_ps();
 171         fix1             = _mm_setzero_ps();
 172         fiy1             = _mm_setzero_ps();
 173         fiz1             = _mm_setzero_ps();
 174         fix2             = _mm_setzero_ps();
 175         fiy2             = _mm_setzero_ps();
 176         fiz2             = _mm_setzero_ps();
 177         fix3             = _mm_setzero_ps();
 178         fiy3             = _mm_setzero_ps();
 179         fiz3             = _mm_setzero_ps();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm_setzero_ps();
 183         vvdwsum          = _mm_setzero_ps();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               x+j_coord_offsetC,x+j_coord_offsetD,
 202                                               &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm_sub_ps(ix0,jx0);
 206             dy00             = _mm_sub_ps(iy0,jy0);
 207             dz00             = _mm_sub_ps(iz0,jz0);
 208             dx10             = _mm_sub_ps(ix1,jx0);
 209             dy10             = _mm_sub_ps(iy1,jy0);
 210             dz10             = _mm_sub_ps(iz1,jz0);
 211             dx20             = _mm_sub_ps(ix2,jx0);
 212             dy20             = _mm_sub_ps(iy2,jy0);
 213             dz20             = _mm_sub_ps(iz2,jz0);
 214             dx30             = _mm_sub_ps(ix3,jx0);
 215             dy30             = _mm_sub_ps(iy3,jy0);
 216             dz30             = _mm_sub_ps(iz3,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 220             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 221             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 222             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 223
 224             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 226             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 227
 228             rinvsq00         = gmx_mm_inv_ps(rsq00);
 229             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 230             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 231             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 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             /* Compute parameters for interactions between i and j atoms */
 250             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 251                                          vdwparam+vdwioffset0+vdwjidx0B,
 252                                          vdwparam+vdwioffset0+vdwjidx0C,
 253                                          vdwparam+vdwioffset0+vdwjidx0D,
 254                                          &c6_00,&c12_00);
 255
 256             /* LENNARD-JONES DISPERSION/REPULSION */
 257
 258             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 259             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 260             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 261             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 262             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 266
 267             fscal            = fvdw;
 268
 269              /* Update vectorial force */
 270             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 271             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 272             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 273
 274             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 275             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 276             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             qq10             = _mm_mul_ps(iq1,jq0);
 284
 285             /* COULOMB ELECTROSTATICS */
 286             velec            = _mm_mul_ps(qq10,rinv10);
 287             felec            = _mm_mul_ps(velec,rinvsq10);
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             velecsum         = _mm_add_ps(velecsum,velec);
 291
 292             fscal            = felec;
 293
 294              /* Update vectorial force */
 295             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 296             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 297             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 298
 299             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 300             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 301             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq20             = _mm_mul_ps(iq2,jq0);
 309
 310             /* COULOMB ELECTROSTATICS */
 311             velec            = _mm_mul_ps(qq20,rinv20);
 312             felec            = _mm_mul_ps(velec,rinvsq20);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velecsum         = _mm_add_ps(velecsum,velec);
 316
 317             fscal            = felec;
 318
 319              /* Update vectorial force */
 320             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 321             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 322             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 323
 324             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 325             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 326             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq30             = _mm_mul_ps(iq3,jq0);
 334
 335             /* COULOMB ELECTROSTATICS */
 336             velec            = _mm_mul_ps(qq30,rinv30);
 337             felec            = _mm_mul_ps(velec,rinvsq30);
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             velecsum         = _mm_add_ps(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344              /* Update vectorial force */
 345             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 346             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 347             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 348
 349             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 350             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 351             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 352
 353             fjptrA             = f+j_coord_offsetA;
 354             fjptrB             = f+j_coord_offsetB;
 355             fjptrC             = f+j_coord_offsetC;
 356             fjptrD             = f+j_coord_offsetD;
 357
 358             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 359
 360             /* Inner loop uses 128 flops */
 361         }
 362
 363         if(jidx<j_index_end)
 364         {
 365
 366             /* Get j neighbor index, and coordinate index */
 367             jnrlistA         = jjnr[jidx];
 368             jnrlistB         = jjnr[jidx+1];
 369             jnrlistC         = jjnr[jidx+2];
 370             jnrlistD         = jjnr[jidx+3];
 371             /* Sign of each element will be negative for non-real atoms.
 372              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 373              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 374              */
 375             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 376             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 377             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 378             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 379             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 380             j_coord_offsetA  = DIM*jnrA;
 381             j_coord_offsetB  = DIM*jnrB;
 382             j_coord_offsetC  = DIM*jnrC;
 383             j_coord_offsetD  = DIM*jnrD;
 384
 385             /* load j atom coordinates */
 386             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 387                                               x+j_coord_offsetC,x+j_coord_offsetD,
 388                                               &jx0,&jy0,&jz0);
 389
 390             /* Calculate displacement vector */
 391             dx00             = _mm_sub_ps(ix0,jx0);
 392             dy00             = _mm_sub_ps(iy0,jy0);
 393             dz00             = _mm_sub_ps(iz0,jz0);
 394             dx10             = _mm_sub_ps(ix1,jx0);
 395             dy10             = _mm_sub_ps(iy1,jy0);
 396             dz10             = _mm_sub_ps(iz1,jz0);
 397             dx20             = _mm_sub_ps(ix2,jx0);
 398             dy20             = _mm_sub_ps(iy2,jy0);
 399             dz20             = _mm_sub_ps(iz2,jz0);
 400             dx30             = _mm_sub_ps(ix3,jx0);
 401             dy30             = _mm_sub_ps(iy3,jy0);
 402             dz30             = _mm_sub_ps(iz3,jz0);
 403
 404             /* Calculate squared distance and things based on it */
 405             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 406             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 407             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 408             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 409
 410             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 411             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 412             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 413
 414             rinvsq00         = gmx_mm_inv_ps(rsq00);
 415             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 416             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 417             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 418
 419             /* Load parameters for j particles */
 420             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 421                                                               charge+jnrC+0,charge+jnrD+0);
 422             vdwjidx0A        = 2*vdwtype[jnrA+0];
 423             vdwjidx0B        = 2*vdwtype[jnrB+0];
 424             vdwjidx0C        = 2*vdwtype[jnrC+0];
 425             vdwjidx0D        = 2*vdwtype[jnrD+0];
 426
 427             fjx0             = _mm_setzero_ps();
 428             fjy0             = _mm_setzero_ps();
 429             fjz0             = _mm_setzero_ps();
 430
 431             /**************************
 432              * CALCULATE INTERACTIONS *
 433              **************************/
 434
 435             /* Compute parameters for interactions between i and j atoms */
 436             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 437                                          vdwparam+vdwioffset0+vdwjidx0B,
 438                                          vdwparam+vdwioffset0+vdwjidx0C,
 439                                          vdwparam+vdwioffset0+vdwjidx0D,
 440                                          &c6_00,&c12_00);
 441
 442             /* LENNARD-JONES DISPERSION/REPULSION */
 443
 444             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 445             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 446             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 447             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 448             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 452             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 453
 454             fscal            = fvdw;
 455
 456             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 457
 458              /* Update vectorial force */
 459             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 460             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 461             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 462
 463             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 464             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 465             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             qq10             = _mm_mul_ps(iq1,jq0);
 473
 474             /* COULOMB ELECTROSTATICS */
 475             velec            = _mm_mul_ps(qq10,rinv10);
 476             felec            = _mm_mul_ps(velec,rinvsq10);
 477
 478             /* Update potential sum for this i atom from the interaction with this j atom. */
 479             velec            = _mm_andnot_ps(dummy_mask,velec);
 480             velecsum         = _mm_add_ps(velecsum,velec);
 481
 482             fscal            = felec;
 483
 484             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 485
 486              /* Update vectorial force */
 487             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 488             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 489             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 490
 491             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 492             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 493             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq20             = _mm_mul_ps(iq2,jq0);
 501
 502             /* COULOMB ELECTROSTATICS */
 503             velec            = _mm_mul_ps(qq20,rinv20);
 504             felec            = _mm_mul_ps(velec,rinvsq20);
 505
 506             /* Update potential sum for this i atom from the interaction with this j atom. */
 507             velec            = _mm_andnot_ps(dummy_mask,velec);
 508             velecsum         = _mm_add_ps(velecsum,velec);
 509
 510             fscal            = felec;
 511
 512             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 513
 514              /* Update vectorial force */
 515             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 516             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 517             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 518
 519             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 520             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 521             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             /* Compute parameters for interactions between i and j atoms */
 528             qq30             = _mm_mul_ps(iq3,jq0);
 529
 530             /* COULOMB ELECTROSTATICS */
 531             velec            = _mm_mul_ps(qq30,rinv30);
 532             felec            = _mm_mul_ps(velec,rinvsq30);
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm_andnot_ps(dummy_mask,velec);
 536             velecsum         = _mm_add_ps(velecsum,velec);
 537
 538             fscal            = felec;
 539
 540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 541
 542              /* Update vectorial force */
 543             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 544             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 545             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 546
 547             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 548             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 549             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 550
 551             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 552             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 553             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 554             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 555
 556             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 557
 558             /* Inner loop uses 128 flops */
 559         }
 560
 561         /* End of innermost loop */
 562
 563         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 564                                               f+i_coord_offset,fshift+i_shift_offset);
 565
 566         ggid                        = gid[iidx];
 567         /* Update potential energies */
 568         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 569         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 570
 571         /* Increment number of inner iterations */
 572         inneriter                  += j_index_end - j_index_start;
 573
 574         /* Outer loop uses 26 flops */
 575     }
 576
 577     /* Increment number of outer iterations */
 578     outeriter        += nri;
 579
 580     /* Update outer/inner flops */
 581
 582     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*128);
 583 }
 584 /*
 585  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_128_fma_single
 586  * Electrostatics interaction: Coulomb
 587  * VdW interaction:            LennardJones
 588  * Geometry:                   Water4-Particle
 589  * Calculate force/pot:        Force
 590  */
 591 void
 592 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_128_fma_single
 593                     (t_nblist                    * gmx_restrict       nlist,
 594                      rvec                        * gmx_restrict          xx,
 595                      rvec                        * gmx_restrict          ff,
 596                      t_forcerec                  * gmx_restrict          fr,
 597                      t_mdatoms                   * gmx_restrict     mdatoms,
 598                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 599                      t_nrnb                      * gmx_restrict        nrnb)
 600 {
 601     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 602      * just 0 for non-waters.
 603      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 604      * jnr indices corresponding to data put in the four positions in the SIMD register.
 605      */
 606     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 607     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 608     int              jnrA,jnrB,jnrC,jnrD;
 609     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 610     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 611     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 612     real             rcutoff_scalar;
 613     real             *shiftvec,*fshift,*x,*f;
 614     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 615     real             scratch[4*DIM];
 616     __m128           fscal,rcutoff,rcutoff2,jidxall;
 617     int              vdwioffset0;
 618     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 619     int              vdwioffset1;
 620     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 621     int              vdwioffset2;
 622     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 623     int              vdwioffset3;
 624     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 625     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 626     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 627     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 628     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 629     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 630     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 631     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 632     real             *charge;
 633     int              nvdwtype;
 634     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 635     int              *vdwtype;
 636     real             *vdwparam;
 637     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 638     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 639     __m128           dummy_mask,cutoff_mask;
 640     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 641     __m128           one     = _mm_set1_ps(1.0);
 642     __m128           two     = _mm_set1_ps(2.0);
 643     x                = xx[0];
 644     f                = ff[0];
 645
 646     nri              = nlist->nri;
 647     iinr             = nlist->iinr;
 648     jindex           = nlist->jindex;
 649     jjnr             = nlist->jjnr;
 650     shiftidx         = nlist->shift;
 651     gid              = nlist->gid;
 652     shiftvec         = fr->shift_vec[0];
 653     fshift           = fr->fshift[0];
 654     facel            = _mm_set1_ps(fr->epsfac);
 655     charge           = mdatoms->chargeA;
 656     nvdwtype         = fr->ntype;
 657     vdwparam         = fr->nbfp;
 658     vdwtype          = mdatoms->typeA;
 659
 660     /* Setup water-specific parameters */
 661     inr              = nlist->iinr[0];
 662     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 663     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 664     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 665     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 666
 667     /* Avoid stupid compiler warnings */
 668     jnrA = jnrB = jnrC = jnrD = 0;
 669     j_coord_offsetA = 0;
 670     j_coord_offsetB = 0;
 671     j_coord_offsetC = 0;
 672     j_coord_offsetD = 0;
 673
 674     outeriter        = 0;
 675     inneriter        = 0;
 676
 677     for(iidx=0;iidx<4*DIM;iidx++)
 678     {
 679         scratch[iidx] = 0.0;
 680     }
 681
 682     /* Start outer loop over neighborlists */
 683     for(iidx=0; iidx<nri; iidx++)
 684     {
 685         /* Load shift vector for this list */
 686         i_shift_offset   = DIM*shiftidx[iidx];
 687
 688         /* Load limits for loop over neighbors */
 689         j_index_start    = jindex[iidx];
 690         j_index_end      = jindex[iidx+1];
 691
 692         /* Get outer coordinate index */
 693         inr              = iinr[iidx];
 694         i_coord_offset   = DIM*inr;
 695
 696         /* Load i particle coords and add shift vector */
 697         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 698                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 699
 700         fix0             = _mm_setzero_ps();
 701         fiy0             = _mm_setzero_ps();
 702         fiz0             = _mm_setzero_ps();
 703         fix1             = _mm_setzero_ps();
 704         fiy1             = _mm_setzero_ps();
 705         fiz1             = _mm_setzero_ps();
 706         fix2             = _mm_setzero_ps();
 707         fiy2             = _mm_setzero_ps();
 708         fiz2             = _mm_setzero_ps();
 709         fix3             = _mm_setzero_ps();
 710         fiy3             = _mm_setzero_ps();
 711         fiz3             = _mm_setzero_ps();
 712
 713         /* Start inner kernel loop */
 714         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 715         {
 716
 717             /* Get j neighbor index, and coordinate index */
 718             jnrA             = jjnr[jidx];
 719             jnrB             = jjnr[jidx+1];
 720             jnrC             = jjnr[jidx+2];
 721             jnrD             = jjnr[jidx+3];
 722             j_coord_offsetA  = DIM*jnrA;
 723             j_coord_offsetB  = DIM*jnrB;
 724             j_coord_offsetC  = DIM*jnrC;
 725             j_coord_offsetD  = DIM*jnrD;
 726
 727             /* load j atom coordinates */
 728             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 729                                               x+j_coord_offsetC,x+j_coord_offsetD,
 730                                               &jx0,&jy0,&jz0);
 731
 732             /* Calculate displacement vector */
 733             dx00             = _mm_sub_ps(ix0,jx0);
 734             dy00             = _mm_sub_ps(iy0,jy0);
 735             dz00             = _mm_sub_ps(iz0,jz0);
 736             dx10             = _mm_sub_ps(ix1,jx0);
 737             dy10             = _mm_sub_ps(iy1,jy0);
 738             dz10             = _mm_sub_ps(iz1,jz0);
 739             dx20             = _mm_sub_ps(ix2,jx0);
 740             dy20             = _mm_sub_ps(iy2,jy0);
 741             dz20             = _mm_sub_ps(iz2,jz0);
 742             dx30             = _mm_sub_ps(ix3,jx0);
 743             dy30             = _mm_sub_ps(iy3,jy0);
 744             dz30             = _mm_sub_ps(iz3,jz0);
 745
 746             /* Calculate squared distance and things based on it */
 747             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 748             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 749             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 750             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 751
 752             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 753             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 754             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 755
 756             rinvsq00         = gmx_mm_inv_ps(rsq00);
 757             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 758             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 759             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 760
 761             /* Load parameters for j particles */
 762             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 763                                                               charge+jnrC+0,charge+jnrD+0);
 764             vdwjidx0A        = 2*vdwtype[jnrA+0];
 765             vdwjidx0B        = 2*vdwtype[jnrB+0];
 766             vdwjidx0C        = 2*vdwtype[jnrC+0];
 767             vdwjidx0D        = 2*vdwtype[jnrD+0];
 768
 769             fjx0             = _mm_setzero_ps();
 770             fjy0             = _mm_setzero_ps();
 771             fjz0             = _mm_setzero_ps();
 772
 773             /**************************
 774              * CALCULATE INTERACTIONS *
 775              **************************/
 776
 777             /* Compute parameters for interactions between i and j atoms */
 778             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 779                                          vdwparam+vdwioffset0+vdwjidx0B,
 780                                          vdwparam+vdwioffset0+vdwjidx0C,
 781                                          vdwparam+vdwioffset0+vdwjidx0D,
 782                                          &c6_00,&c12_00);
 783
 784             /* LENNARD-JONES DISPERSION/REPULSION */
 785
 786             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 787             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 788
 789             fscal            = fvdw;
 790
 791              /* Update vectorial force */
 792             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 793             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 794             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 795
 796             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 797             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 798             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 799
 800             /**************************
 801              * CALCULATE INTERACTIONS *
 802              **************************/
 803
 804             /* Compute parameters for interactions between i and j atoms */
 805             qq10             = _mm_mul_ps(iq1,jq0);
 806
 807             /* COULOMB ELECTROSTATICS */
 808             velec            = _mm_mul_ps(qq10,rinv10);
 809             felec            = _mm_mul_ps(velec,rinvsq10);
 810
 811             fscal            = felec;
 812
 813              /* Update vectorial force */
 814             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 815             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 816             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 817
 818             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 819             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 820             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 821
 822             /**************************
 823              * CALCULATE INTERACTIONS *
 824              **************************/
 825
 826             /* Compute parameters for interactions between i and j atoms */
 827             qq20             = _mm_mul_ps(iq2,jq0);
 828
 829             /* COULOMB ELECTROSTATICS */
 830             velec            = _mm_mul_ps(qq20,rinv20);
 831             felec            = _mm_mul_ps(velec,rinvsq20);
 832
 833             fscal            = felec;
 834
 835              /* Update vectorial force */
 836             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 837             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 838             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 839
 840             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 841             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 842             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 843
 844             /**************************
 845              * CALCULATE INTERACTIONS *
 846              **************************/
 847
 848             /* Compute parameters for interactions between i and j atoms */
 849             qq30             = _mm_mul_ps(iq3,jq0);
 850
 851             /* COULOMB ELECTROSTATICS */
 852             velec            = _mm_mul_ps(qq30,rinv30);
 853             felec            = _mm_mul_ps(velec,rinvsq30);
 854
 855             fscal            = felec;
 856
 857              /* Update vectorial force */
 858             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 859             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 860             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 861
 862             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 863             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 864             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 865
 866             fjptrA             = f+j_coord_offsetA;
 867             fjptrB             = f+j_coord_offsetB;
 868             fjptrC             = f+j_coord_offsetC;
 869             fjptrD             = f+j_coord_offsetD;
 870
 871             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 872
 873             /* Inner loop uses 120 flops */
 874         }
 875
 876         if(jidx<j_index_end)
 877         {
 878
 879             /* Get j neighbor index, and coordinate index */
 880             jnrlistA         = jjnr[jidx];
 881             jnrlistB         = jjnr[jidx+1];
 882             jnrlistC         = jjnr[jidx+2];
 883             jnrlistD         = jjnr[jidx+3];
 884             /* Sign of each element will be negative for non-real atoms.
 885              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 886              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 887              */
 888             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 889             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 890             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 891             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 892             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 893             j_coord_offsetA  = DIM*jnrA;
 894             j_coord_offsetB  = DIM*jnrB;
 895             j_coord_offsetC  = DIM*jnrC;
 896             j_coord_offsetD  = DIM*jnrD;
 897
 898             /* load j atom coordinates */
 899             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 900                                               x+j_coord_offsetC,x+j_coord_offsetD,
 901                                               &jx0,&jy0,&jz0);
 902
 903             /* Calculate displacement vector */
 904             dx00             = _mm_sub_ps(ix0,jx0);
 905             dy00             = _mm_sub_ps(iy0,jy0);
 906             dz00             = _mm_sub_ps(iz0,jz0);
 907             dx10             = _mm_sub_ps(ix1,jx0);
 908             dy10             = _mm_sub_ps(iy1,jy0);
 909             dz10             = _mm_sub_ps(iz1,jz0);
 910             dx20             = _mm_sub_ps(ix2,jx0);
 911             dy20             = _mm_sub_ps(iy2,jy0);
 912             dz20             = _mm_sub_ps(iz2,jz0);
 913             dx30             = _mm_sub_ps(ix3,jx0);
 914             dy30             = _mm_sub_ps(iy3,jy0);
 915             dz30             = _mm_sub_ps(iz3,jz0);
 916
 917             /* Calculate squared distance and things based on it */
 918             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 919             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 920             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 921             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 922
 923             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 924             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 925             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 926
 927             rinvsq00         = gmx_mm_inv_ps(rsq00);
 928             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 929             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 930             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 931
 932             /* Load parameters for j particles */
 933             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 934                                                               charge+jnrC+0,charge+jnrD+0);
 935             vdwjidx0A        = 2*vdwtype[jnrA+0];
 936             vdwjidx0B        = 2*vdwtype[jnrB+0];
 937             vdwjidx0C        = 2*vdwtype[jnrC+0];
 938             vdwjidx0D        = 2*vdwtype[jnrD+0];
 939
 940             fjx0             = _mm_setzero_ps();
 941             fjy0             = _mm_setzero_ps();
 942             fjz0             = _mm_setzero_ps();
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             /* Compute parameters for interactions between i and j atoms */
 949             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 950                                          vdwparam+vdwioffset0+vdwjidx0B,
 951                                          vdwparam+vdwioffset0+vdwjidx0C,
 952                                          vdwparam+vdwioffset0+vdwjidx0D,
 953                                          &c6_00,&c12_00);
 954
 955             /* LENNARD-JONES DISPERSION/REPULSION */
 956
 957             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 958             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 959
 960             fscal            = fvdw;
 961
 962             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 963
 964              /* Update vectorial force */
 965             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 966             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 967             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 968
 969             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 970             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 971             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 972
 973             /**************************
 974              * CALCULATE INTERACTIONS *
 975              **************************/
 976
 977             /* Compute parameters for interactions between i and j atoms */
 978             qq10             = _mm_mul_ps(iq1,jq0);
 979
 980             /* COULOMB ELECTROSTATICS */
 981             velec            = _mm_mul_ps(qq10,rinv10);
 982             felec            = _mm_mul_ps(velec,rinvsq10);
 983
 984             fscal            = felec;
 985
 986             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 987
 988              /* Update vectorial force */
 989             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 990             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 991             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 992
 993             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 994             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 995             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             /* Compute parameters for interactions between i and j atoms */
1002             qq20             = _mm_mul_ps(iq2,jq0);
1003
1004             /* COULOMB ELECTROSTATICS */
1005             velec            = _mm_mul_ps(qq20,rinv20);
1006             felec            = _mm_mul_ps(velec,rinvsq20);
1007
1008             fscal            = felec;
1009
1010             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1011
1012              /* Update vectorial force */
1013             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1014             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1015             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1016
1017             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1018             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1019             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             /* Compute parameters for interactions between i and j atoms */
1026             qq30             = _mm_mul_ps(iq3,jq0);
1027
1028             /* COULOMB ELECTROSTATICS */
1029             velec            = _mm_mul_ps(qq30,rinv30);
1030             felec            = _mm_mul_ps(velec,rinvsq30);
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1035
1036              /* Update vectorial force */
1037             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1038             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1039             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1040
1041             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1042             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1043             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1044
1045             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1046             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1047             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1048             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1049
1050             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1051
1052             /* Inner loop uses 120 flops */
1053         }
1054
1055         /* End of innermost loop */
1056
1057         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1058                                               f+i_coord_offset,fshift+i_shift_offset);
1059
1060         /* Increment number of inner iterations */
1061         inneriter                  += j_index_end - j_index_start;
1062
1063         /* Outer loop uses 24 flops */
1064     }
1065
1066     /* Increment number of outer iterations */
1067     outeriter        += nri;
1068
1069     /* Update outer/inner flops */
1070
1071     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*120);
1072 }