src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_avx_128_fma_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "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_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_avx_128_fma_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128d           c6grid_00;
 102     __m128d           c6grid_10;
 103     __m128d           c6grid_20;
 104     __m128d           c6grid_30;
 105     real             *vdwgridparam;
 106     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 107     __m128d           one_half  = _mm_set1_pd(0.5);
 108     __m128d           minus_one = _mm_set1_pd(-1.0);
 109     __m128i          ewitab;
 110     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     real             *ewtab;
 112     __m128d          dummy_mask,cutoff_mask;
 113     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 114     __m128d          one     = _mm_set1_pd(1.0);
 115     __m128d          two     = _mm_set1_pd(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm_set1_pd(fr->ic->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132     vdwgridparam     = fr->ljpme_c6grid;
 133     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 134     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 135     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 136
 137     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 138     ewtab            = fr->ic->tabq_coul_FDV0;
 139     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 140     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 145     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 146     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 147     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 150     rcutoff_scalar   = fr->ic->rcoulomb;
 151     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 152     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 153
 154     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 155     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 156
 157     /* Avoid stupid compiler warnings */
 158     jnrA = jnrB = 0;
 159     j_coord_offsetA = 0;
 160     j_coord_offsetB = 0;
 161
 162     outeriter        = 0;
 163     inneriter        = 0;
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 182
 183         fix0             = _mm_setzero_pd();
 184         fiy0             = _mm_setzero_pd();
 185         fiz0             = _mm_setzero_pd();
 186         fix1             = _mm_setzero_pd();
 187         fiy1             = _mm_setzero_pd();
 188         fiz1             = _mm_setzero_pd();
 189         fix2             = _mm_setzero_pd();
 190         fiy2             = _mm_setzero_pd();
 191         fiz2             = _mm_setzero_pd();
 192         fix3             = _mm_setzero_pd();
 193         fiy3             = _mm_setzero_pd();
 194         fiz3             = _mm_setzero_pd();
 195
 196         /* Reset potential sums */
 197         velecsum         = _mm_setzero_pd();
 198         vvdwsum          = _mm_setzero_pd();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209
 210             /* load j atom coordinates */
 211             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 212                                               &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm_sub_pd(ix0,jx0);
 216             dy00             = _mm_sub_pd(iy0,jy0);
 217             dz00             = _mm_sub_pd(iz0,jz0);
 218             dx10             = _mm_sub_pd(ix1,jx0);
 219             dy10             = _mm_sub_pd(iy1,jy0);
 220             dz10             = _mm_sub_pd(iz1,jz0);
 221             dx20             = _mm_sub_pd(ix2,jx0);
 222             dy20             = _mm_sub_pd(iy2,jy0);
 223             dz20             = _mm_sub_pd(iz2,jz0);
 224             dx30             = _mm_sub_pd(ix3,jx0);
 225             dy30             = _mm_sub_pd(iy3,jy0);
 226             dz30             = _mm_sub_pd(iz3,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 230             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 231             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 232             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 233
 234             rinv00           = avx128fma_invsqrt_d(rsq00);
 235             rinv10           = avx128fma_invsqrt_d(rsq10);
 236             rinv20           = avx128fma_invsqrt_d(rsq20);
 237             rinv30           = avx128fma_invsqrt_d(rsq30);
 238
 239             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 240             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 241             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 242             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 243
 244             /* Load parameters for j particles */
 245             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 246             vdwjidx0A        = 2*vdwtype[jnrA+0];
 247             vdwjidx0B        = 2*vdwtype[jnrB+0];
 248
 249             fjx0             = _mm_setzero_pd();
 250             fjy0             = _mm_setzero_pd();
 251             fjz0             = _mm_setzero_pd();
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             if (gmx_mm_any_lt(rsq00,rcutoff2))
 258             {
 259
 260             r00              = _mm_mul_pd(rsq00,rinv00);
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 264                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 265             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 266                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 267
 268             /* Analytical LJ-PME */
 269             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 270             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 271             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 272             exponent         = avx128fma_exp_d(ewcljrsq);
 273             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 274             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 275             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 276             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 277             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 278             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 279                                _mm_mul_pd(_mm_sub_pd(vvdw6,_mm_macc_pd(c6grid_00,sh_lj_ewald,_mm_mul_pd(c6_00,sh_vdw_invrcut6))),one_sixth));
 280             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 281             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 282
 283             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 284
 285             /* Update potential sum for this i atom from the interaction with this j atom. */
 286             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 287             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 288
 289             fscal            = fvdw;
 290
 291             fscal            = _mm_and_pd(fscal,cutoff_mask);
 292
 293             /* Update vectorial force */
 294             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 295             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 296             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 297
 298             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 299             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 300             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 301
 302             }
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             if (gmx_mm_any_lt(rsq10,rcutoff2))
 309             {
 310
 311             r10              = _mm_mul_pd(rsq10,rinv10);
 312
 313             /* Compute parameters for interactions between i and j atoms */
 314             qq10             = _mm_mul_pd(iq1,jq0);
 315
 316             /* EWALD ELECTROSTATICS */
 317
 318             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 319             ewrt             = _mm_mul_pd(r10,ewtabscale);
 320             ewitab           = _mm_cvttpd_epi32(ewrt);
 321 #ifdef __XOP__
 322             eweps            = _mm_frcz_pd(ewrt);
 323 #else
 324             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 325 #endif
 326             twoeweps         = _mm_add_pd(eweps,eweps);
 327             ewitab           = _mm_slli_epi32(ewitab,2);
 328             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 329             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 330             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 331             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 332             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 333             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 334             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 335             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 336             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 337             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 338
 339             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velec            = _mm_and_pd(velec,cutoff_mask);
 343             velecsum         = _mm_add_pd(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             fscal            = _mm_and_pd(fscal,cutoff_mask);
 348
 349             /* Update vectorial force */
 350             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 351             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 352             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 353
 354             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 355             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 356             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 357
 358             }
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             if (gmx_mm_any_lt(rsq20,rcutoff2))
 365             {
 366
 367             r20              = _mm_mul_pd(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm_mul_pd(iq2,jq0);
 371
 372             /* EWALD ELECTROSTATICS */
 373
 374             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 375             ewrt             = _mm_mul_pd(r20,ewtabscale);
 376             ewitab           = _mm_cvttpd_epi32(ewrt);
 377 #ifdef __XOP__
 378             eweps            = _mm_frcz_pd(ewrt);
 379 #else
 380             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 381 #endif
 382             twoeweps         = _mm_add_pd(eweps,eweps);
 383             ewitab           = _mm_slli_epi32(ewitab,2);
 384             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 385             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 386             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 387             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 388             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 389             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 390             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 391             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 392             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 393             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 394
 395             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 396
 397             /* Update potential sum for this i atom from the interaction with this j atom. */
 398             velec            = _mm_and_pd(velec,cutoff_mask);
 399             velecsum         = _mm_add_pd(velecsum,velec);
 400
 401             fscal            = felec;
 402
 403             fscal            = _mm_and_pd(fscal,cutoff_mask);
 404
 405             /* Update vectorial force */
 406             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 407             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 408             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 409
 410             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 411             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 412             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 413
 414             }
 415
 416             /**************************
 417              * CALCULATE INTERACTIONS *
 418              **************************/
 419
 420             if (gmx_mm_any_lt(rsq30,rcutoff2))
 421             {
 422
 423             r30              = _mm_mul_pd(rsq30,rinv30);
 424
 425             /* Compute parameters for interactions between i and j atoms */
 426             qq30             = _mm_mul_pd(iq3,jq0);
 427
 428             /* EWALD ELECTROSTATICS */
 429
 430             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 431             ewrt             = _mm_mul_pd(r30,ewtabscale);
 432             ewitab           = _mm_cvttpd_epi32(ewrt);
 433 #ifdef __XOP__
 434             eweps            = _mm_frcz_pd(ewrt);
 435 #else
 436             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 437 #endif
 438             twoeweps         = _mm_add_pd(eweps,eweps);
 439             ewitab           = _mm_slli_epi32(ewitab,2);
 440             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 441             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 442             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 443             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 444             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 445             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 446             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 447             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 448             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 449             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 450
 451             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 452
 453             /* Update potential sum for this i atom from the interaction with this j atom. */
 454             velec            = _mm_and_pd(velec,cutoff_mask);
 455             velecsum         = _mm_add_pd(velecsum,velec);
 456
 457             fscal            = felec;
 458
 459             fscal            = _mm_and_pd(fscal,cutoff_mask);
 460
 461             /* Update vectorial force */
 462             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 463             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 464             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 465
 466             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 467             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 468             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 469
 470             }
 471
 472             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 473
 474             /* Inner loop uses 208 flops */
 475         }
 476
 477         if(jidx<j_index_end)
 478         {
 479
 480             jnrA             = jjnr[jidx];
 481             j_coord_offsetA  = DIM*jnrA;
 482
 483             /* load j atom coordinates */
 484             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 485                                               &jx0,&jy0,&jz0);
 486
 487             /* Calculate displacement vector */
 488             dx00             = _mm_sub_pd(ix0,jx0);
 489             dy00             = _mm_sub_pd(iy0,jy0);
 490             dz00             = _mm_sub_pd(iz0,jz0);
 491             dx10             = _mm_sub_pd(ix1,jx0);
 492             dy10             = _mm_sub_pd(iy1,jy0);
 493             dz10             = _mm_sub_pd(iz1,jz0);
 494             dx20             = _mm_sub_pd(ix2,jx0);
 495             dy20             = _mm_sub_pd(iy2,jy0);
 496             dz20             = _mm_sub_pd(iz2,jz0);
 497             dx30             = _mm_sub_pd(ix3,jx0);
 498             dy30             = _mm_sub_pd(iy3,jy0);
 499             dz30             = _mm_sub_pd(iz3,jz0);
 500
 501             /* Calculate squared distance and things based on it */
 502             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 503             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 504             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 505             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 506
 507             rinv00           = avx128fma_invsqrt_d(rsq00);
 508             rinv10           = avx128fma_invsqrt_d(rsq10);
 509             rinv20           = avx128fma_invsqrt_d(rsq20);
 510             rinv30           = avx128fma_invsqrt_d(rsq30);
 511
 512             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 513             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 514             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 515             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 516
 517             /* Load parameters for j particles */
 518             jq0              = _mm_load_sd(charge+jnrA+0);
 519             vdwjidx0A        = 2*vdwtype[jnrA+0];
 520
 521             fjx0             = _mm_setzero_pd();
 522             fjy0             = _mm_setzero_pd();
 523             fjz0             = _mm_setzero_pd();
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             if (gmx_mm_any_lt(rsq00,rcutoff2))
 530             {
 531
 532             r00              = _mm_mul_pd(rsq00,rinv00);
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 536             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 537
 538             /* Analytical LJ-PME */
 539             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 540             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 541             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 542             exponent         = avx128fma_exp_d(ewcljrsq);
 543             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 544             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 545             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 546             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 547             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 548             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 549                                _mm_mul_pd(_mm_sub_pd(vvdw6,_mm_macc_pd(c6grid_00,sh_lj_ewald,_mm_mul_pd(c6_00,sh_vdw_invrcut6))),one_sixth));
 550             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 551             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 552
 553             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 554
 555             /* Update potential sum for this i atom from the interaction with this j atom. */
 556             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 557             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 558             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 559
 560             fscal            = fvdw;
 561
 562             fscal            = _mm_and_pd(fscal,cutoff_mask);
 563
 564             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 565
 566             /* Update vectorial force */
 567             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 568             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 569             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 570
 571             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 572             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 573             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 574
 575             }
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             if (gmx_mm_any_lt(rsq10,rcutoff2))
 582             {
 583
 584             r10              = _mm_mul_pd(rsq10,rinv10);
 585
 586             /* Compute parameters for interactions between i and j atoms */
 587             qq10             = _mm_mul_pd(iq1,jq0);
 588
 589             /* EWALD ELECTROSTATICS */
 590
 591             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 592             ewrt             = _mm_mul_pd(r10,ewtabscale);
 593             ewitab           = _mm_cvttpd_epi32(ewrt);
 594 #ifdef __XOP__
 595             eweps            = _mm_frcz_pd(ewrt);
 596 #else
 597             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 598 #endif
 599             twoeweps         = _mm_add_pd(eweps,eweps);
 600             ewitab           = _mm_slli_epi32(ewitab,2);
 601             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 602             ewtabD           = _mm_setzero_pd();
 603             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 604             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 605             ewtabFn          = _mm_setzero_pd();
 606             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 607             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 608             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 609             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 610             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 611
 612             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velec            = _mm_and_pd(velec,cutoff_mask);
 616             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 617             velecsum         = _mm_add_pd(velecsum,velec);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm_and_pd(fscal,cutoff_mask);
 622
 623             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 624
 625             /* Update vectorial force */
 626             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 627             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 628             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 629
 630             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 631             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 632             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 633
 634             }
 635
 636             /**************************
 637              * CALCULATE INTERACTIONS *
 638              **************************/
 639
 640             if (gmx_mm_any_lt(rsq20,rcutoff2))
 641             {
 642
 643             r20              = _mm_mul_pd(rsq20,rinv20);
 644
 645             /* Compute parameters for interactions between i and j atoms */
 646             qq20             = _mm_mul_pd(iq2,jq0);
 647
 648             /* EWALD ELECTROSTATICS */
 649
 650             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 651             ewrt             = _mm_mul_pd(r20,ewtabscale);
 652             ewitab           = _mm_cvttpd_epi32(ewrt);
 653 #ifdef __XOP__
 654             eweps            = _mm_frcz_pd(ewrt);
 655 #else
 656             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 657 #endif
 658             twoeweps         = _mm_add_pd(eweps,eweps);
 659             ewitab           = _mm_slli_epi32(ewitab,2);
 660             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 661             ewtabD           = _mm_setzero_pd();
 662             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 663             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 664             ewtabFn          = _mm_setzero_pd();
 665             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 666             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 667             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 668             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 669             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 670
 671             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 672
 673             /* Update potential sum for this i atom from the interaction with this j atom. */
 674             velec            = _mm_and_pd(velec,cutoff_mask);
 675             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 676             velecsum         = _mm_add_pd(velecsum,velec);
 677
 678             fscal            = felec;
 679
 680             fscal            = _mm_and_pd(fscal,cutoff_mask);
 681
 682             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 683
 684             /* Update vectorial force */
 685             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 686             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 687             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 688
 689             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 690             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 691             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 692
 693             }
 694
 695             /**************************
 696              * CALCULATE INTERACTIONS *
 697              **************************/
 698
 699             if (gmx_mm_any_lt(rsq30,rcutoff2))
 700             {
 701
 702             r30              = _mm_mul_pd(rsq30,rinv30);
 703
 704             /* Compute parameters for interactions between i and j atoms */
 705             qq30             = _mm_mul_pd(iq3,jq0);
 706
 707             /* EWALD ELECTROSTATICS */
 708
 709             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 710             ewrt             = _mm_mul_pd(r30,ewtabscale);
 711             ewitab           = _mm_cvttpd_epi32(ewrt);
 712 #ifdef __XOP__
 713             eweps            = _mm_frcz_pd(ewrt);
 714 #else
 715             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 716 #endif
 717             twoeweps         = _mm_add_pd(eweps,eweps);
 718             ewitab           = _mm_slli_epi32(ewitab,2);
 719             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 720             ewtabD           = _mm_setzero_pd();
 721             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 722             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 723             ewtabFn          = _mm_setzero_pd();
 724             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 725             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 726             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 727             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 728             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 729
 730             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 731
 732             /* Update potential sum for this i atom from the interaction with this j atom. */
 733             velec            = _mm_and_pd(velec,cutoff_mask);
 734             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 735             velecsum         = _mm_add_pd(velecsum,velec);
 736
 737             fscal            = felec;
 738
 739             fscal            = _mm_and_pd(fscal,cutoff_mask);
 740
 741             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 742
 743             /* Update vectorial force */
 744             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 745             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 746             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 747
 748             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 749             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 750             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 751
 752             }
 753
 754             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 755
 756             /* Inner loop uses 208 flops */
 757         }
 758
 759         /* End of innermost loop */
 760
 761         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 762                                               f+i_coord_offset,fshift+i_shift_offset);
 763
 764         ggid                        = gid[iidx];
 765         /* Update potential energies */
 766         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 767         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 768
 769         /* Increment number of inner iterations */
 770         inneriter                  += j_index_end - j_index_start;
 771
 772         /* Outer loop uses 26 flops */
 773     }
 774
 775     /* Increment number of outer iterations */
 776     outeriter        += nri;
 777
 778     /* Update outer/inner flops */
 779
 780     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*208);
 781 }
 782 /*
 783  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_128_fma_double
 784  * Electrostatics interaction: Ewald
 785  * VdW interaction:            LJEwald
 786  * Geometry:                   Water4-Particle
 787  * Calculate force/pot:        Force
 788  */
 789 void
 790 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_128_fma_double
 791                     (t_nblist                    * gmx_restrict       nlist,
 792                      rvec                        * gmx_restrict          xx,
 793                      rvec                        * gmx_restrict          ff,
 794                      struct t_forcerec           * gmx_restrict          fr,
 795                      t_mdatoms                   * gmx_restrict     mdatoms,
 796                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 797                      t_nrnb                      * gmx_restrict        nrnb)
 798 {
 799     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 800      * just 0 for non-waters.
 801      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 802      * jnr indices corresponding to data put in the four positions in the SIMD register.
 803      */
 804     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 805     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 806     int              jnrA,jnrB;
 807     int              j_coord_offsetA,j_coord_offsetB;
 808     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 809     real             rcutoff_scalar;
 810     real             *shiftvec,*fshift,*x,*f;
 811     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 812     int              vdwioffset0;
 813     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 814     int              vdwioffset1;
 815     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 816     int              vdwioffset2;
 817     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 818     int              vdwioffset3;
 819     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 820     int              vdwjidx0A,vdwjidx0B;
 821     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 822     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 823     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 824     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 825     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 826     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 827     real             *charge;
 828     int              nvdwtype;
 829     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 830     int              *vdwtype;
 831     real             *vdwparam;
 832     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 833     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 834     __m128d           c6grid_00;
 835     __m128d           c6grid_10;
 836     __m128d           c6grid_20;
 837     __m128d           c6grid_30;
 838     real             *vdwgridparam;
 839     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 840     __m128d           one_half  = _mm_set1_pd(0.5);
 841     __m128d           minus_one = _mm_set1_pd(-1.0);
 842     __m128i          ewitab;
 843     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 844     real             *ewtab;
 845     __m128d          dummy_mask,cutoff_mask;
 846     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 847     __m128d          one     = _mm_set1_pd(1.0);
 848     __m128d          two     = _mm_set1_pd(2.0);
 849     x                = xx[0];
 850     f                = ff[0];
 851
 852     nri              = nlist->nri;
 853     iinr             = nlist->iinr;
 854     jindex           = nlist->jindex;
 855     jjnr             = nlist->jjnr;
 856     shiftidx         = nlist->shift;
 857     gid              = nlist->gid;
 858     shiftvec         = fr->shift_vec[0];
 859     fshift           = fr->fshift[0];
 860     facel            = _mm_set1_pd(fr->ic->epsfac);
 861     charge           = mdatoms->chargeA;
 862     nvdwtype         = fr->ntype;
 863     vdwparam         = fr->nbfp;
 864     vdwtype          = mdatoms->typeA;
 865     vdwgridparam     = fr->ljpme_c6grid;
 866     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 867     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 868     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 869
 870     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 871     ewtab            = fr->ic->tabq_coul_F;
 872     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 873     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 874
 875     /* Setup water-specific parameters */
 876     inr              = nlist->iinr[0];
 877     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 878     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 879     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 880     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 881
 882     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 883     rcutoff_scalar   = fr->ic->rcoulomb;
 884     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 885     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 886
 887     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 888     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 889
 890     /* Avoid stupid compiler warnings */
 891     jnrA = jnrB = 0;
 892     j_coord_offsetA = 0;
 893     j_coord_offsetB = 0;
 894
 895     outeriter        = 0;
 896     inneriter        = 0;
 897
 898     /* Start outer loop over neighborlists */
 899     for(iidx=0; iidx<nri; iidx++)
 900     {
 901         /* Load shift vector for this list */
 902         i_shift_offset   = DIM*shiftidx[iidx];
 903
 904         /* Load limits for loop over neighbors */
 905         j_index_start    = jindex[iidx];
 906         j_index_end      = jindex[iidx+1];
 907
 908         /* Get outer coordinate index */
 909         inr              = iinr[iidx];
 910         i_coord_offset   = DIM*inr;
 911
 912         /* Load i particle coords and add shift vector */
 913         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 914                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 915
 916         fix0             = _mm_setzero_pd();
 917         fiy0             = _mm_setzero_pd();
 918         fiz0             = _mm_setzero_pd();
 919         fix1             = _mm_setzero_pd();
 920         fiy1             = _mm_setzero_pd();
 921         fiz1             = _mm_setzero_pd();
 922         fix2             = _mm_setzero_pd();
 923         fiy2             = _mm_setzero_pd();
 924         fiz2             = _mm_setzero_pd();
 925         fix3             = _mm_setzero_pd();
 926         fiy3             = _mm_setzero_pd();
 927         fiz3             = _mm_setzero_pd();
 928
 929         /* Start inner kernel loop */
 930         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 931         {
 932
 933             /* Get j neighbor index, and coordinate index */
 934             jnrA             = jjnr[jidx];
 935             jnrB             = jjnr[jidx+1];
 936             j_coord_offsetA  = DIM*jnrA;
 937             j_coord_offsetB  = DIM*jnrB;
 938
 939             /* load j atom coordinates */
 940             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 941                                               &jx0,&jy0,&jz0);
 942
 943             /* Calculate displacement vector */
 944             dx00             = _mm_sub_pd(ix0,jx0);
 945             dy00             = _mm_sub_pd(iy0,jy0);
 946             dz00             = _mm_sub_pd(iz0,jz0);
 947             dx10             = _mm_sub_pd(ix1,jx0);
 948             dy10             = _mm_sub_pd(iy1,jy0);
 949             dz10             = _mm_sub_pd(iz1,jz0);
 950             dx20             = _mm_sub_pd(ix2,jx0);
 951             dy20             = _mm_sub_pd(iy2,jy0);
 952             dz20             = _mm_sub_pd(iz2,jz0);
 953             dx30             = _mm_sub_pd(ix3,jx0);
 954             dy30             = _mm_sub_pd(iy3,jy0);
 955             dz30             = _mm_sub_pd(iz3,jz0);
 956
 957             /* Calculate squared distance and things based on it */
 958             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 959             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 960             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 961             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 962
 963             rinv00           = avx128fma_invsqrt_d(rsq00);
 964             rinv10           = avx128fma_invsqrt_d(rsq10);
 965             rinv20           = avx128fma_invsqrt_d(rsq20);
 966             rinv30           = avx128fma_invsqrt_d(rsq30);
 967
 968             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 969             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 970             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 971             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 972
 973             /* Load parameters for j particles */
 974             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 975             vdwjidx0A        = 2*vdwtype[jnrA+0];
 976             vdwjidx0B        = 2*vdwtype[jnrB+0];
 977
 978             fjx0             = _mm_setzero_pd();
 979             fjy0             = _mm_setzero_pd();
 980             fjz0             = _mm_setzero_pd();
 981
 982             /**************************
 983              * CALCULATE INTERACTIONS *
 984              **************************/
 985
 986             if (gmx_mm_any_lt(rsq00,rcutoff2))
 987             {
 988
 989             r00              = _mm_mul_pd(rsq00,rinv00);
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 993                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 994             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 995                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 996
 997             /* Analytical LJ-PME */
 998             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 999             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1000             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1001             exponent         = avx128fma_exp_d(ewcljrsq);
1002             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1003             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1004             /* f6A = 6 * C6grid * (1 - poly) */
1005             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1006             /* f6B = C6grid * exponent * beta^6 */
1007             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1008             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1009             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1010
1011             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1012
1013             fscal            = fvdw;
1014
1015             fscal            = _mm_and_pd(fscal,cutoff_mask);
1016
1017             /* Update vectorial force */
1018             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1019             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1020             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1021
1022             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1023             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1024             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1025
1026             }
1027
1028             /**************************
1029              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             if (gmx_mm_any_lt(rsq10,rcutoff2))
1033             {
1034
1035             r10              = _mm_mul_pd(rsq10,rinv10);
1036
1037             /* Compute parameters for interactions between i and j atoms */
1038             qq10             = _mm_mul_pd(iq1,jq0);
1039
1040             /* EWALD ELECTROSTATICS */
1041
1042             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1043             ewrt             = _mm_mul_pd(r10,ewtabscale);
1044             ewitab           = _mm_cvttpd_epi32(ewrt);
1045 #ifdef __XOP__
1046             eweps            = _mm_frcz_pd(ewrt);
1047 #else
1048             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1049 #endif
1050             twoeweps         = _mm_add_pd(eweps,eweps);
1051             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1052                                          &ewtabF,&ewtabFn);
1053             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1054             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1055
1056             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1057
1058             fscal            = felec;
1059
1060             fscal            = _mm_and_pd(fscal,cutoff_mask);
1061
1062             /* Update vectorial force */
1063             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1064             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1065             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1066
1067             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1068             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1069             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1070
1071             }
1072
1073             /**************************
1074              * CALCULATE INTERACTIONS *
1075              **************************/
1076
1077             if (gmx_mm_any_lt(rsq20,rcutoff2))
1078             {
1079
1080             r20              = _mm_mul_pd(rsq20,rinv20);
1081
1082             /* Compute parameters for interactions between i and j atoms */
1083             qq20             = _mm_mul_pd(iq2,jq0);
1084
1085             /* EWALD ELECTROSTATICS */
1086
1087             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1088             ewrt             = _mm_mul_pd(r20,ewtabscale);
1089             ewitab           = _mm_cvttpd_epi32(ewrt);
1090 #ifdef __XOP__
1091             eweps            = _mm_frcz_pd(ewrt);
1092 #else
1093             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1094 #endif
1095             twoeweps         = _mm_add_pd(eweps,eweps);
1096             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1097                                          &ewtabF,&ewtabFn);
1098             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1099             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1100
1101             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1102
1103             fscal            = felec;
1104
1105             fscal            = _mm_and_pd(fscal,cutoff_mask);
1106
1107             /* Update vectorial force */
1108             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1109             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1110             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1111
1112             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1113             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1114             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1115
1116             }
1117
1118             /**************************
1119              * CALCULATE INTERACTIONS *
1120              **************************/
1121
1122             if (gmx_mm_any_lt(rsq30,rcutoff2))
1123             {
1124
1125             r30              = _mm_mul_pd(rsq30,rinv30);
1126
1127             /* Compute parameters for interactions between i and j atoms */
1128             qq30             = _mm_mul_pd(iq3,jq0);
1129
1130             /* EWALD ELECTROSTATICS */
1131
1132             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1133             ewrt             = _mm_mul_pd(r30,ewtabscale);
1134             ewitab           = _mm_cvttpd_epi32(ewrt);
1135 #ifdef __XOP__
1136             eweps            = _mm_frcz_pd(ewrt);
1137 #else
1138             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1139 #endif
1140             twoeweps         = _mm_add_pd(eweps,eweps);
1141             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1142                                          &ewtabF,&ewtabFn);
1143             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1144             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1145
1146             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1147
1148             fscal            = felec;
1149
1150             fscal            = _mm_and_pd(fscal,cutoff_mask);
1151
1152             /* Update vectorial force */
1153             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1154             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1155             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1156
1157             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1158             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1159             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1160
1161             }
1162
1163             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1164
1165             /* Inner loop uses 179 flops */
1166         }
1167
1168         if(jidx<j_index_end)
1169         {
1170
1171             jnrA             = jjnr[jidx];
1172             j_coord_offsetA  = DIM*jnrA;
1173
1174             /* load j atom coordinates */
1175             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1176                                               &jx0,&jy0,&jz0);
1177
1178             /* Calculate displacement vector */
1179             dx00             = _mm_sub_pd(ix0,jx0);
1180             dy00             = _mm_sub_pd(iy0,jy0);
1181             dz00             = _mm_sub_pd(iz0,jz0);
1182             dx10             = _mm_sub_pd(ix1,jx0);
1183             dy10             = _mm_sub_pd(iy1,jy0);
1184             dz10             = _mm_sub_pd(iz1,jz0);
1185             dx20             = _mm_sub_pd(ix2,jx0);
1186             dy20             = _mm_sub_pd(iy2,jy0);
1187             dz20             = _mm_sub_pd(iz2,jz0);
1188             dx30             = _mm_sub_pd(ix3,jx0);
1189             dy30             = _mm_sub_pd(iy3,jy0);
1190             dz30             = _mm_sub_pd(iz3,jz0);
1191
1192             /* Calculate squared distance and things based on it */
1193             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1194             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1195             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1196             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1197
1198             rinv00           = avx128fma_invsqrt_d(rsq00);
1199             rinv10           = avx128fma_invsqrt_d(rsq10);
1200             rinv20           = avx128fma_invsqrt_d(rsq20);
1201             rinv30           = avx128fma_invsqrt_d(rsq30);
1202
1203             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1204             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1205             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1206             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1207
1208             /* Load parameters for j particles */
1209             jq0              = _mm_load_sd(charge+jnrA+0);
1210             vdwjidx0A        = 2*vdwtype[jnrA+0];
1211
1212             fjx0             = _mm_setzero_pd();
1213             fjy0             = _mm_setzero_pd();
1214             fjz0             = _mm_setzero_pd();
1215
1216             /**************************
1217              * CALCULATE INTERACTIONS *
1218              **************************/
1219
1220             if (gmx_mm_any_lt(rsq00,rcutoff2))
1221             {
1222
1223             r00              = _mm_mul_pd(rsq00,rinv00);
1224
1225             /* Compute parameters for interactions between i and j atoms */
1226             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1227             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1228
1229             /* Analytical LJ-PME */
1230             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1231             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1232             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1233             exponent         = avx128fma_exp_d(ewcljrsq);
1234             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1235             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1236             /* f6A = 6 * C6grid * (1 - poly) */
1237             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1238             /* f6B = C6grid * exponent * beta^6 */
1239             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1240             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1241             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1242
1243             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1244
1245             fscal            = fvdw;
1246
1247             fscal            = _mm_and_pd(fscal,cutoff_mask);
1248
1249             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1250
1251             /* Update vectorial force */
1252             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1253             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1254             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1255
1256             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1257             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1258             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1259
1260             }
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             if (gmx_mm_any_lt(rsq10,rcutoff2))
1267             {
1268
1269             r10              = _mm_mul_pd(rsq10,rinv10);
1270
1271             /* Compute parameters for interactions between i and j atoms */
1272             qq10             = _mm_mul_pd(iq1,jq0);
1273
1274             /* EWALD ELECTROSTATICS */
1275
1276             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1277             ewrt             = _mm_mul_pd(r10,ewtabscale);
1278             ewitab           = _mm_cvttpd_epi32(ewrt);
1279 #ifdef __XOP__
1280             eweps            = _mm_frcz_pd(ewrt);
1281 #else
1282             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1283 #endif
1284             twoeweps         = _mm_add_pd(eweps,eweps);
1285             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1286             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1287             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1288
1289             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1290
1291             fscal            = felec;
1292
1293             fscal            = _mm_and_pd(fscal,cutoff_mask);
1294
1295             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1296
1297             /* Update vectorial force */
1298             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1299             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1300             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1301
1302             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1303             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1304             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1305
1306             }
1307
1308             /**************************
1309              * CALCULATE INTERACTIONS *
1310              **************************/
1311
1312             if (gmx_mm_any_lt(rsq20,rcutoff2))
1313             {
1314
1315             r20              = _mm_mul_pd(rsq20,rinv20);
1316
1317             /* Compute parameters for interactions between i and j atoms */
1318             qq20             = _mm_mul_pd(iq2,jq0);
1319
1320             /* EWALD ELECTROSTATICS */
1321
1322             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1323             ewrt             = _mm_mul_pd(r20,ewtabscale);
1324             ewitab           = _mm_cvttpd_epi32(ewrt);
1325 #ifdef __XOP__
1326             eweps            = _mm_frcz_pd(ewrt);
1327 #else
1328             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1329 #endif
1330             twoeweps         = _mm_add_pd(eweps,eweps);
1331             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1332             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1333             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1334
1335             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1336
1337             fscal            = felec;
1338
1339             fscal            = _mm_and_pd(fscal,cutoff_mask);
1340
1341             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1342
1343             /* Update vectorial force */
1344             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1345             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1346             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1347
1348             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1349             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1350             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1351
1352             }
1353
1354             /**************************
1355              * CALCULATE INTERACTIONS *
1356              **************************/
1357
1358             if (gmx_mm_any_lt(rsq30,rcutoff2))
1359             {
1360
1361             r30              = _mm_mul_pd(rsq30,rinv30);
1362
1363             /* Compute parameters for interactions between i and j atoms */
1364             qq30             = _mm_mul_pd(iq3,jq0);
1365
1366             /* EWALD ELECTROSTATICS */
1367
1368             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1369             ewrt             = _mm_mul_pd(r30,ewtabscale);
1370             ewitab           = _mm_cvttpd_epi32(ewrt);
1371 #ifdef __XOP__
1372             eweps            = _mm_frcz_pd(ewrt);
1373 #else
1374             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1375 #endif
1376             twoeweps         = _mm_add_pd(eweps,eweps);
1377             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1378             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1379             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1380
1381             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1382
1383             fscal            = felec;
1384
1385             fscal            = _mm_and_pd(fscal,cutoff_mask);
1386
1387             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1388
1389             /* Update vectorial force */
1390             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1391             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1392             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1393
1394             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1395             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1396             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1397
1398             }
1399
1400             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1401
1402             /* Inner loop uses 179 flops */
1403         }
1404
1405         /* End of innermost loop */
1406
1407         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1408                                               f+i_coord_offset,fshift+i_shift_offset);
1409
1410         /* Increment number of inner iterations */
1411         inneriter                  += j_index_end - j_index_start;
1412
1413         /* Outer loop uses 24 flops */
1414     }
1415
1416     /* Increment number of outer iterations */
1417     outeriter        += nri;
1418
1419     /* Update outer/inner flops */
1420
1421     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*179);
1422 }