src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwLJ_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128i          ewitab;
 102     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 103     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 104     real             *ewtab;
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_ps(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 127     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 128     beta2            = _mm_mul_ps(beta,beta);
 129     beta3            = _mm_mul_ps(beta,beta2);
 130     ewtab            = fr->ic->tabq_coul_FDV0;
 131     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 132     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 137     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 138     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 139     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     for(iidx=0;iidx<4*DIM;iidx++)
 152     {
 153         scratch[iidx] = 0.0;
 154     }
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 173
 174         fix0             = _mm_setzero_ps();
 175         fiy0             = _mm_setzero_ps();
 176         fiz0             = _mm_setzero_ps();
 177         fix1             = _mm_setzero_ps();
 178         fiy1             = _mm_setzero_ps();
 179         fiz1             = _mm_setzero_ps();
 180         fix2             = _mm_setzero_ps();
 181         fiy2             = _mm_setzero_ps();
 182         fiz2             = _mm_setzero_ps();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_ps();
 186         vvdwsum          = _mm_setzero_ps();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                               x+j_coord_offsetC,x+j_coord_offsetD,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_ps(ix0,jx0);
 209             dy00             = _mm_sub_ps(iy0,jy0);
 210             dz00             = _mm_sub_ps(iz0,jz0);
 211             dx10             = _mm_sub_ps(ix1,jx0);
 212             dy10             = _mm_sub_ps(iy1,jy0);
 213             dz10             = _mm_sub_ps(iz1,jz0);
 214             dx20             = _mm_sub_ps(ix2,jx0);
 215             dy20             = _mm_sub_ps(iy2,jy0);
 216             dz20             = _mm_sub_ps(iz2,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
 223             rinv00           = avx128fma_invsqrt_f(rsq00);
 224             rinv10           = avx128fma_invsqrt_f(rsq10);
 225             rinv20           = avx128fma_invsqrt_f(rsq20);
 226
 227             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 228             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                               charge+jnrC+0,charge+jnrD+0);
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235             vdwjidx0B        = 2*vdwtype[jnrB+0];
 236             vdwjidx0C        = 2*vdwtype[jnrC+0];
 237             vdwjidx0D        = 2*vdwtype[jnrD+0];
 238
 239             fjx0             = _mm_setzero_ps();
 240             fjy0             = _mm_setzero_ps();
 241             fjz0             = _mm_setzero_ps();
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             r00              = _mm_mul_ps(rsq00,rinv00);
 248
 249             /* Compute parameters for interactions between i and j atoms */
 250             qq00             = _mm_mul_ps(iq0,jq0);
 251             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,
 253                                          vdwparam+vdwioffset0+vdwjidx0C,
 254                                          vdwparam+vdwioffset0+vdwjidx0D,
 255                                          &c6_00,&c12_00);
 256
 257             /* EWALD ELECTROSTATICS */
 258
 259             /* Analytical PME correction */
 260             zeta2            = _mm_mul_ps(beta2,rsq00);
 261             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 262             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 263             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 264             felec            = _mm_mul_ps(qq00,felec);
 265             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 266             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 267             velec            = _mm_mul_ps(qq00,velec);
 268
 269             /* LENNARD-JONES DISPERSION/REPULSION */
 270
 271             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 272             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 273             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 274             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 275             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velecsum         = _mm_add_ps(velecsum,velec);
 279             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 280
 281             fscal            = _mm_add_ps(felec,fvdw);
 282
 283              /* Update vectorial force */
 284             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 285             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 286             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 287
 288             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 289             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 290             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             r10              = _mm_mul_ps(rsq10,rinv10);
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq10             = _mm_mul_ps(iq1,jq0);
 300
 301             /* EWALD ELECTROSTATICS */
 302
 303             /* Analytical PME correction */
 304             zeta2            = _mm_mul_ps(beta2,rsq10);
 305             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 306             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 307             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 308             felec            = _mm_mul_ps(qq10,felec);
 309             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 310             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 311             velec            = _mm_mul_ps(qq10,velec);
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velecsum         = _mm_add_ps(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318              /* Update vectorial force */
 319             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 320             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 321             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 322
 323             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 324             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 325             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             r20              = _mm_mul_ps(rsq20,rinv20);
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             qq20             = _mm_mul_ps(iq2,jq0);
 335
 336             /* EWALD ELECTROSTATICS */
 337
 338             /* Analytical PME correction */
 339             zeta2            = _mm_mul_ps(beta2,rsq20);
 340             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 341             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 342             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 343             felec            = _mm_mul_ps(qq20,felec);
 344             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 345             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 346             velec            = _mm_mul_ps(qq20,velec);
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353              /* Update vectorial force */
 354             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 355             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 356             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 357
 358             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 359             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 360             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 361
 362             fjptrA             = f+j_coord_offsetA;
 363             fjptrB             = f+j_coord_offsetB;
 364             fjptrC             = f+j_coord_offsetC;
 365             fjptrD             = f+j_coord_offsetD;
 366
 367             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 368
 369             /* Inner loop uses 99 flops */
 370         }
 371
 372         if(jidx<j_index_end)
 373         {
 374
 375             /* Get j neighbor index, and coordinate index */
 376             jnrlistA         = jjnr[jidx];
 377             jnrlistB         = jjnr[jidx+1];
 378             jnrlistC         = jjnr[jidx+2];
 379             jnrlistD         = jjnr[jidx+3];
 380             /* Sign of each element will be negative for non-real atoms.
 381              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 382              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 383              */
 384             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 385             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 386             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 387             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 388             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 389             j_coord_offsetA  = DIM*jnrA;
 390             j_coord_offsetB  = DIM*jnrB;
 391             j_coord_offsetC  = DIM*jnrC;
 392             j_coord_offsetD  = DIM*jnrD;
 393
 394             /* load j atom coordinates */
 395             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 396                                               x+j_coord_offsetC,x+j_coord_offsetD,
 397                                               &jx0,&jy0,&jz0);
 398
 399             /* Calculate displacement vector */
 400             dx00             = _mm_sub_ps(ix0,jx0);
 401             dy00             = _mm_sub_ps(iy0,jy0);
 402             dz00             = _mm_sub_ps(iz0,jz0);
 403             dx10             = _mm_sub_ps(ix1,jx0);
 404             dy10             = _mm_sub_ps(iy1,jy0);
 405             dz10             = _mm_sub_ps(iz1,jz0);
 406             dx20             = _mm_sub_ps(ix2,jx0);
 407             dy20             = _mm_sub_ps(iy2,jy0);
 408             dz20             = _mm_sub_ps(iz2,jz0);
 409
 410             /* Calculate squared distance and things based on it */
 411             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 412             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 413             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 414
 415             rinv00           = avx128fma_invsqrt_f(rsq00);
 416             rinv10           = avx128fma_invsqrt_f(rsq10);
 417             rinv20           = avx128fma_invsqrt_f(rsq20);
 418
 419             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 420             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 421             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 422
 423             /* Load parameters for j particles */
 424             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 425                                                               charge+jnrC+0,charge+jnrD+0);
 426             vdwjidx0A        = 2*vdwtype[jnrA+0];
 427             vdwjidx0B        = 2*vdwtype[jnrB+0];
 428             vdwjidx0C        = 2*vdwtype[jnrC+0];
 429             vdwjidx0D        = 2*vdwtype[jnrD+0];
 430
 431             fjx0             = _mm_setzero_ps();
 432             fjy0             = _mm_setzero_ps();
 433             fjz0             = _mm_setzero_ps();
 434
 435             /**************************
 436              * CALCULATE INTERACTIONS *
 437              **************************/
 438
 439             r00              = _mm_mul_ps(rsq00,rinv00);
 440             r00              = _mm_andnot_ps(dummy_mask,r00);
 441
 442             /* Compute parameters for interactions between i and j atoms */
 443             qq00             = _mm_mul_ps(iq0,jq0);
 444             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 445                                          vdwparam+vdwioffset0+vdwjidx0B,
 446                                          vdwparam+vdwioffset0+vdwjidx0C,
 447                                          vdwparam+vdwioffset0+vdwjidx0D,
 448                                          &c6_00,&c12_00);
 449
 450             /* EWALD ELECTROSTATICS */
 451
 452             /* Analytical PME correction */
 453             zeta2            = _mm_mul_ps(beta2,rsq00);
 454             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 455             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 456             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 457             felec            = _mm_mul_ps(qq00,felec);
 458             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 459             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 460             velec            = _mm_mul_ps(qq00,velec);
 461
 462             /* LENNARD-JONES DISPERSION/REPULSION */
 463
 464             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 465             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 466             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 467             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 468             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 469
 470             /* Update potential sum for this i atom from the interaction with this j atom. */
 471             velec            = _mm_andnot_ps(dummy_mask,velec);
 472             velecsum         = _mm_add_ps(velecsum,velec);
 473             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 474             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 475
 476             fscal            = _mm_add_ps(felec,fvdw);
 477
 478             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 479
 480              /* Update vectorial force */
 481             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 482             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 483             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 484
 485             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 486             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 487             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             r10              = _mm_mul_ps(rsq10,rinv10);
 494             r10              = _mm_andnot_ps(dummy_mask,r10);
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             qq10             = _mm_mul_ps(iq1,jq0);
 498
 499             /* EWALD ELECTROSTATICS */
 500
 501             /* Analytical PME correction */
 502             zeta2            = _mm_mul_ps(beta2,rsq10);
 503             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 504             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 505             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 506             felec            = _mm_mul_ps(qq10,felec);
 507             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 508             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 509             velec            = _mm_mul_ps(qq10,velec);
 510
 511             /* Update potential sum for this i atom from the interaction with this j atom. */
 512             velec            = _mm_andnot_ps(dummy_mask,velec);
 513             velecsum         = _mm_add_ps(velecsum,velec);
 514
 515             fscal            = felec;
 516
 517             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 518
 519              /* Update vectorial force */
 520             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 521             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 522             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 523
 524             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 525             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 526             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             r20              = _mm_mul_ps(rsq20,rinv20);
 533             r20              = _mm_andnot_ps(dummy_mask,r20);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq20             = _mm_mul_ps(iq2,jq0);
 537
 538             /* EWALD ELECTROSTATICS */
 539
 540             /* Analytical PME correction */
 541             zeta2            = _mm_mul_ps(beta2,rsq20);
 542             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 543             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 544             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 545             felec            = _mm_mul_ps(qq20,felec);
 546             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 547             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 548             velec            = _mm_mul_ps(qq20,velec);
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _mm_andnot_ps(dummy_mask,velec);
 552             velecsum         = _mm_add_ps(velecsum,velec);
 553
 554             fscal            = felec;
 555
 556             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 557
 558              /* Update vectorial force */
 559             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 560             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 561             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 562
 563             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 564             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 565             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 566
 567             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 568             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 569             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 570             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 571
 572             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 573
 574             /* Inner loop uses 102 flops */
 575         }
 576
 577         /* End of innermost loop */
 578
 579         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 580                                               f+i_coord_offset,fshift+i_shift_offset);
 581
 582         ggid                        = gid[iidx];
 583         /* Update potential energies */
 584         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 585         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 586
 587         /* Increment number of inner iterations */
 588         inneriter                  += j_index_end - j_index_start;
 589
 590         /* Outer loop uses 20 flops */
 591     }
 592
 593     /* Increment number of outer iterations */
 594     outeriter        += nri;
 595
 596     /* Update outer/inner flops */
 597
 598     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*102);
 599 }
 600 /*
 601  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_single
 602  * Electrostatics interaction: Ewald
 603  * VdW interaction:            LennardJones
 604  * Geometry:                   Water3-Particle
 605  * Calculate force/pot:        Force
 606  */
 607 void
 608 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_single
 609                     (t_nblist                    * gmx_restrict       nlist,
 610                      rvec                        * gmx_restrict          xx,
 611                      rvec                        * gmx_restrict          ff,
 612                      struct t_forcerec           * gmx_restrict          fr,
 613                      t_mdatoms                   * gmx_restrict     mdatoms,
 614                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 615                      t_nrnb                      * gmx_restrict        nrnb)
 616 {
 617     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 618      * just 0 for non-waters.
 619      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 620      * jnr indices corresponding to data put in the four positions in the SIMD register.
 621      */
 622     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 623     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 624     int              jnrA,jnrB,jnrC,jnrD;
 625     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 626     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 627     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 628     real             rcutoff_scalar;
 629     real             *shiftvec,*fshift,*x,*f;
 630     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 631     real             scratch[4*DIM];
 632     __m128           fscal,rcutoff,rcutoff2,jidxall;
 633     int              vdwioffset0;
 634     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 635     int              vdwioffset1;
 636     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 637     int              vdwioffset2;
 638     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 639     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 640     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 641     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 642     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 643     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 644     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 645     real             *charge;
 646     int              nvdwtype;
 647     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 648     int              *vdwtype;
 649     real             *vdwparam;
 650     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 651     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 652     __m128i          ewitab;
 653     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 654     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 655     real             *ewtab;
 656     __m128           dummy_mask,cutoff_mask;
 657     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 658     __m128           one     = _mm_set1_ps(1.0);
 659     __m128           two     = _mm_set1_ps(2.0);
 660     x                = xx[0];
 661     f                = ff[0];
 662
 663     nri              = nlist->nri;
 664     iinr             = nlist->iinr;
 665     jindex           = nlist->jindex;
 666     jjnr             = nlist->jjnr;
 667     shiftidx         = nlist->shift;
 668     gid              = nlist->gid;
 669     shiftvec         = fr->shift_vec[0];
 670     fshift           = fr->fshift[0];
 671     facel            = _mm_set1_ps(fr->ic->epsfac);
 672     charge           = mdatoms->chargeA;
 673     nvdwtype         = fr->ntype;
 674     vdwparam         = fr->nbfp;
 675     vdwtype          = mdatoms->typeA;
 676
 677     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 678     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 679     beta2            = _mm_mul_ps(beta,beta);
 680     beta3            = _mm_mul_ps(beta,beta2);
 681     ewtab            = fr->ic->tabq_coul_F;
 682     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 683     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 684
 685     /* Setup water-specific parameters */
 686     inr              = nlist->iinr[0];
 687     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 688     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 689     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 690     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 691
 692     /* Avoid stupid compiler warnings */
 693     jnrA = jnrB = jnrC = jnrD = 0;
 694     j_coord_offsetA = 0;
 695     j_coord_offsetB = 0;
 696     j_coord_offsetC = 0;
 697     j_coord_offsetD = 0;
 698
 699     outeriter        = 0;
 700     inneriter        = 0;
 701
 702     for(iidx=0;iidx<4*DIM;iidx++)
 703     {
 704         scratch[iidx] = 0.0;
 705     }
 706
 707     /* Start outer loop over neighborlists */
 708     for(iidx=0; iidx<nri; iidx++)
 709     {
 710         /* Load shift vector for this list */
 711         i_shift_offset   = DIM*shiftidx[iidx];
 712
 713         /* Load limits for loop over neighbors */
 714         j_index_start    = jindex[iidx];
 715         j_index_end      = jindex[iidx+1];
 716
 717         /* Get outer coordinate index */
 718         inr              = iinr[iidx];
 719         i_coord_offset   = DIM*inr;
 720
 721         /* Load i particle coords and add shift vector */
 722         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 723                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 724
 725         fix0             = _mm_setzero_ps();
 726         fiy0             = _mm_setzero_ps();
 727         fiz0             = _mm_setzero_ps();
 728         fix1             = _mm_setzero_ps();
 729         fiy1             = _mm_setzero_ps();
 730         fiz1             = _mm_setzero_ps();
 731         fix2             = _mm_setzero_ps();
 732         fiy2             = _mm_setzero_ps();
 733         fiz2             = _mm_setzero_ps();
 734
 735         /* Start inner kernel loop */
 736         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 737         {
 738
 739             /* Get j neighbor index, and coordinate index */
 740             jnrA             = jjnr[jidx];
 741             jnrB             = jjnr[jidx+1];
 742             jnrC             = jjnr[jidx+2];
 743             jnrD             = jjnr[jidx+3];
 744             j_coord_offsetA  = DIM*jnrA;
 745             j_coord_offsetB  = DIM*jnrB;
 746             j_coord_offsetC  = DIM*jnrC;
 747             j_coord_offsetD  = DIM*jnrD;
 748
 749             /* load j atom coordinates */
 750             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 751                                               x+j_coord_offsetC,x+j_coord_offsetD,
 752                                               &jx0,&jy0,&jz0);
 753
 754             /* Calculate displacement vector */
 755             dx00             = _mm_sub_ps(ix0,jx0);
 756             dy00             = _mm_sub_ps(iy0,jy0);
 757             dz00             = _mm_sub_ps(iz0,jz0);
 758             dx10             = _mm_sub_ps(ix1,jx0);
 759             dy10             = _mm_sub_ps(iy1,jy0);
 760             dz10             = _mm_sub_ps(iz1,jz0);
 761             dx20             = _mm_sub_ps(ix2,jx0);
 762             dy20             = _mm_sub_ps(iy2,jy0);
 763             dz20             = _mm_sub_ps(iz2,jz0);
 764
 765             /* Calculate squared distance and things based on it */
 766             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 767             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 768             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 769
 770             rinv00           = avx128fma_invsqrt_f(rsq00);
 771             rinv10           = avx128fma_invsqrt_f(rsq10);
 772             rinv20           = avx128fma_invsqrt_f(rsq20);
 773
 774             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 775             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 776             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 777
 778             /* Load parameters for j particles */
 779             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 780                                                               charge+jnrC+0,charge+jnrD+0);
 781             vdwjidx0A        = 2*vdwtype[jnrA+0];
 782             vdwjidx0B        = 2*vdwtype[jnrB+0];
 783             vdwjidx0C        = 2*vdwtype[jnrC+0];
 784             vdwjidx0D        = 2*vdwtype[jnrD+0];
 785
 786             fjx0             = _mm_setzero_ps();
 787             fjy0             = _mm_setzero_ps();
 788             fjz0             = _mm_setzero_ps();
 789
 790             /**************************
 791              * CALCULATE INTERACTIONS *
 792              **************************/
 793
 794             r00              = _mm_mul_ps(rsq00,rinv00);
 795
 796             /* Compute parameters for interactions between i and j atoms */
 797             qq00             = _mm_mul_ps(iq0,jq0);
 798             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 799                                          vdwparam+vdwioffset0+vdwjidx0B,
 800                                          vdwparam+vdwioffset0+vdwjidx0C,
 801                                          vdwparam+vdwioffset0+vdwjidx0D,
 802                                          &c6_00,&c12_00);
 803
 804             /* EWALD ELECTROSTATICS */
 805
 806             /* Analytical PME correction */
 807             zeta2            = _mm_mul_ps(beta2,rsq00);
 808             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 809             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 810             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 811             felec            = _mm_mul_ps(qq00,felec);
 812
 813             /* LENNARD-JONES DISPERSION/REPULSION */
 814
 815             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 816             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 817
 818             fscal            = _mm_add_ps(felec,fvdw);
 819
 820              /* Update vectorial force */
 821             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 822             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 823             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 824
 825             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 826             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 827             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 828
 829             /**************************
 830              * CALCULATE INTERACTIONS *
 831              **************************/
 832
 833             r10              = _mm_mul_ps(rsq10,rinv10);
 834
 835             /* Compute parameters for interactions between i and j atoms */
 836             qq10             = _mm_mul_ps(iq1,jq0);
 837
 838             /* EWALD ELECTROSTATICS */
 839
 840             /* Analytical PME correction */
 841             zeta2            = _mm_mul_ps(beta2,rsq10);
 842             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 843             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 844             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 845             felec            = _mm_mul_ps(qq10,felec);
 846
 847             fscal            = felec;
 848
 849              /* Update vectorial force */
 850             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 851             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 852             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 853
 854             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 855             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 856             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             r20              = _mm_mul_ps(rsq20,rinv20);
 863
 864             /* Compute parameters for interactions between i and j atoms */
 865             qq20             = _mm_mul_ps(iq2,jq0);
 866
 867             /* EWALD ELECTROSTATICS */
 868
 869             /* Analytical PME correction */
 870             zeta2            = _mm_mul_ps(beta2,rsq20);
 871             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 872             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 873             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 874             felec            = _mm_mul_ps(qq20,felec);
 875
 876             fscal            = felec;
 877
 878              /* Update vectorial force */
 879             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 880             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 881             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 882
 883             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 884             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 885             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 886
 887             fjptrA             = f+j_coord_offsetA;
 888             fjptrB             = f+j_coord_offsetB;
 889             fjptrC             = f+j_coord_offsetC;
 890             fjptrD             = f+j_coord_offsetD;
 891
 892             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 893
 894             /* Inner loop uses 91 flops */
 895         }
 896
 897         if(jidx<j_index_end)
 898         {
 899
 900             /* Get j neighbor index, and coordinate index */
 901             jnrlistA         = jjnr[jidx];
 902             jnrlistB         = jjnr[jidx+1];
 903             jnrlistC         = jjnr[jidx+2];
 904             jnrlistD         = jjnr[jidx+3];
 905             /* Sign of each element will be negative for non-real atoms.
 906              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 907              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 908              */
 909             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 910             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 911             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 912             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 913             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 914             j_coord_offsetA  = DIM*jnrA;
 915             j_coord_offsetB  = DIM*jnrB;
 916             j_coord_offsetC  = DIM*jnrC;
 917             j_coord_offsetD  = DIM*jnrD;
 918
 919             /* load j atom coordinates */
 920             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 921                                               x+j_coord_offsetC,x+j_coord_offsetD,
 922                                               &jx0,&jy0,&jz0);
 923
 924             /* Calculate displacement vector */
 925             dx00             = _mm_sub_ps(ix0,jx0);
 926             dy00             = _mm_sub_ps(iy0,jy0);
 927             dz00             = _mm_sub_ps(iz0,jz0);
 928             dx10             = _mm_sub_ps(ix1,jx0);
 929             dy10             = _mm_sub_ps(iy1,jy0);
 930             dz10             = _mm_sub_ps(iz1,jz0);
 931             dx20             = _mm_sub_ps(ix2,jx0);
 932             dy20             = _mm_sub_ps(iy2,jy0);
 933             dz20             = _mm_sub_ps(iz2,jz0);
 934
 935             /* Calculate squared distance and things based on it */
 936             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 937             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 938             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 939
 940             rinv00           = avx128fma_invsqrt_f(rsq00);
 941             rinv10           = avx128fma_invsqrt_f(rsq10);
 942             rinv20           = avx128fma_invsqrt_f(rsq20);
 943
 944             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 945             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 946             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 947
 948             /* Load parameters for j particles */
 949             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 950                                                               charge+jnrC+0,charge+jnrD+0);
 951             vdwjidx0A        = 2*vdwtype[jnrA+0];
 952             vdwjidx0B        = 2*vdwtype[jnrB+0];
 953             vdwjidx0C        = 2*vdwtype[jnrC+0];
 954             vdwjidx0D        = 2*vdwtype[jnrD+0];
 955
 956             fjx0             = _mm_setzero_ps();
 957             fjy0             = _mm_setzero_ps();
 958             fjz0             = _mm_setzero_ps();
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             r00              = _mm_mul_ps(rsq00,rinv00);
 965             r00              = _mm_andnot_ps(dummy_mask,r00);
 966
 967             /* Compute parameters for interactions between i and j atoms */
 968             qq00             = _mm_mul_ps(iq0,jq0);
 969             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 970                                          vdwparam+vdwioffset0+vdwjidx0B,
 971                                          vdwparam+vdwioffset0+vdwjidx0C,
 972                                          vdwparam+vdwioffset0+vdwjidx0D,
 973                                          &c6_00,&c12_00);
 974
 975             /* EWALD ELECTROSTATICS */
 976
 977             /* Analytical PME correction */
 978             zeta2            = _mm_mul_ps(beta2,rsq00);
 979             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 980             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 981             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 982             felec            = _mm_mul_ps(qq00,felec);
 983
 984             /* LENNARD-JONES DISPERSION/REPULSION */
 985
 986             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 987             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 988
 989             fscal            = _mm_add_ps(felec,fvdw);
 990
 991             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 992
 993              /* Update vectorial force */
 994             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 995             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 996             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 997
 998             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 999             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1000             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             r10              = _mm_mul_ps(rsq10,rinv10);
1007             r10              = _mm_andnot_ps(dummy_mask,r10);
1008
1009             /* Compute parameters for interactions between i and j atoms */
1010             qq10             = _mm_mul_ps(iq1,jq0);
1011
1012             /* EWALD ELECTROSTATICS */
1013
1014             /* Analytical PME correction */
1015             zeta2            = _mm_mul_ps(beta2,rsq10);
1016             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1017             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1018             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1019             felec            = _mm_mul_ps(qq10,felec);
1020
1021             fscal            = felec;
1022
1023             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1024
1025              /* Update vectorial force */
1026             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1027             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1028             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1029
1030             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1031             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1032             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             r20              = _mm_mul_ps(rsq20,rinv20);
1039             r20              = _mm_andnot_ps(dummy_mask,r20);
1040
1041             /* Compute parameters for interactions between i and j atoms */
1042             qq20             = _mm_mul_ps(iq2,jq0);
1043
1044             /* EWALD ELECTROSTATICS */
1045
1046             /* Analytical PME correction */
1047             zeta2            = _mm_mul_ps(beta2,rsq20);
1048             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1049             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1050             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1051             felec            = _mm_mul_ps(qq20,felec);
1052
1053             fscal            = felec;
1054
1055             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1056
1057              /* Update vectorial force */
1058             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1059             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1060             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1061
1062             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1063             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1064             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1065
1066             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1067             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1068             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1069             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1070
1071             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1072
1073             /* Inner loop uses 94 flops */
1074         }
1075
1076         /* End of innermost loop */
1077
1078         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1079                                               f+i_coord_offset,fshift+i_shift_offset);
1080
1081         /* Increment number of inner iterations */
1082         inneriter                  += j_index_end - j_index_start;
1083
1084         /* Outer loop uses 18 flops */
1085     }
1086
1087     /* Increment number of outer iterations */
1088     outeriter        += nri;
1089
1090     /* Update outer/inner flops */
1091
1092     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*94);
1093 }