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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128d           c6grid_00;
 105     __m128d           c6grid_10;
 106     __m128d           c6grid_20;
 107     __m128d           c6grid_30;
 108     real             *vdwgridparam;
 109     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 110     __m128d           one_half  = _mm_set1_pd(0.5);
 111     __m128d           minus_one = _mm_set1_pd(-1.0);
 112     __m128i          ewitab;
 113     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     real             *ewtab;
 115     __m128d          dummy_mask,cutoff_mask;
 116     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 117     __m128d          one     = _mm_set1_pd(1.0);
 118     __m128d          two     = _mm_set1_pd(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_pd(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135     vdwgridparam     = fr->ljpme_c6grid;
 136     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 137     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 138     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 139
 140     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 141     ewtab            = fr->ic->tabq_coul_FDV0;
 142     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 143     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 148     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 149     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 150     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 151
 152     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 153     rcutoff_scalar   = fr->rcoulomb;
 154     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 155     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 156
 157     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 158     rvdw             = _mm_set1_pd(fr->rvdw);
 159
 160     /* Avoid stupid compiler warnings */
 161     jnrA = jnrB = 0;
 162     j_coord_offsetA = 0;
 163     j_coord_offsetB = 0;
 164
 165     outeriter        = 0;
 166     inneriter        = 0;
 167
 168     /* Start outer loop over neighborlists */
 169     for(iidx=0; iidx<nri; iidx++)
 170     {
 171         /* Load shift vector for this list */
 172         i_shift_offset   = DIM*shiftidx[iidx];
 173
 174         /* Load limits for loop over neighbors */
 175         j_index_start    = jindex[iidx];
 176         j_index_end      = jindex[iidx+1];
 177
 178         /* Get outer coordinate index */
 179         inr              = iinr[iidx];
 180         i_coord_offset   = DIM*inr;
 181
 182         /* Load i particle coords and add shift vector */
 183         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 184                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 185
 186         fix0             = _mm_setzero_pd();
 187         fiy0             = _mm_setzero_pd();
 188         fiz0             = _mm_setzero_pd();
 189         fix1             = _mm_setzero_pd();
 190         fiy1             = _mm_setzero_pd();
 191         fiz1             = _mm_setzero_pd();
 192         fix2             = _mm_setzero_pd();
 193         fiy2             = _mm_setzero_pd();
 194         fiz2             = _mm_setzero_pd();
 195         fix3             = _mm_setzero_pd();
 196         fiy3             = _mm_setzero_pd();
 197         fiz3             = _mm_setzero_pd();
 198
 199         /* Reset potential sums */
 200         velecsum         = _mm_setzero_pd();
 201         vvdwsum          = _mm_setzero_pd();
 202
 203         /* Start inner kernel loop */
 204         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 205         {
 206
 207             /* Get j neighbor index, and coordinate index */
 208             jnrA             = jjnr[jidx];
 209             jnrB             = jjnr[jidx+1];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212
 213             /* load j atom coordinates */
 214             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 215                                               &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm_sub_pd(ix0,jx0);
 219             dy00             = _mm_sub_pd(iy0,jy0);
 220             dz00             = _mm_sub_pd(iz0,jz0);
 221             dx10             = _mm_sub_pd(ix1,jx0);
 222             dy10             = _mm_sub_pd(iy1,jy0);
 223             dz10             = _mm_sub_pd(iz1,jz0);
 224             dx20             = _mm_sub_pd(ix2,jx0);
 225             dy20             = _mm_sub_pd(iy2,jy0);
 226             dz20             = _mm_sub_pd(iz2,jz0);
 227             dx30             = _mm_sub_pd(ix3,jx0);
 228             dy30             = _mm_sub_pd(iy3,jy0);
 229             dz30             = _mm_sub_pd(iz3,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 233             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 234             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 235             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 236
 237             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 238             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 239             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 240             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 241
 242             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 243             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 244             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 245             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 246
 247             /* Load parameters for j particles */
 248             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 249             vdwjidx0A        = 2*vdwtype[jnrA+0];
 250             vdwjidx0B        = 2*vdwtype[jnrB+0];
 251
 252             fjx0             = _mm_setzero_pd();
 253             fjy0             = _mm_setzero_pd();
 254             fjz0             = _mm_setzero_pd();
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             if (gmx_mm_any_lt(rsq00,rcutoff2))
 261             {
 262
 263             r00              = _mm_mul_pd(rsq00,rinv00);
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 267                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 268             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 269                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 270
 271             /* Analytical LJ-PME */
 272             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 273             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 274             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 275             exponent         = gmx_simd_exp_d(ewcljrsq);
 276             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 277             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 278             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 279             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 280             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 281             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 282                                _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));
 283             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 284             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);
 285
 286             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 290             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 291
 292             fscal            = fvdw;
 293
 294             fscal            = _mm_and_pd(fscal,cutoff_mask);
 295
 296             /* Update vectorial force */
 297             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 298             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 299             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 300
 301             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 302             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 303             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 304
 305             }
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             if (gmx_mm_any_lt(rsq10,rcutoff2))
 312             {
 313
 314             r10              = _mm_mul_pd(rsq10,rinv10);
 315
 316             /* Compute parameters for interactions between i and j atoms */
 317             qq10             = _mm_mul_pd(iq1,jq0);
 318
 319             /* EWALD ELECTROSTATICS */
 320
 321             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 322             ewrt             = _mm_mul_pd(r10,ewtabscale);
 323             ewitab           = _mm_cvttpd_epi32(ewrt);
 324 #ifdef __XOP__
 325             eweps            = _mm_frcz_pd(ewrt);
 326 #else
 327             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 328 #endif
 329             twoeweps         = _mm_add_pd(eweps,eweps);
 330             ewitab           = _mm_slli_epi32(ewitab,2);
 331             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 332             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 333             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 334             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 335             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 336             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 337             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 338             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 339             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 340             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 341
 342             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velec            = _mm_and_pd(velec,cutoff_mask);
 346             velecsum         = _mm_add_pd(velecsum,velec);
 347
 348             fscal            = felec;
 349
 350             fscal            = _mm_and_pd(fscal,cutoff_mask);
 351
 352             /* Update vectorial force */
 353             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 354             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 355             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 356
 357             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 358             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 359             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 360
 361             }
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             if (gmx_mm_any_lt(rsq20,rcutoff2))
 368             {
 369
 370             r20              = _mm_mul_pd(rsq20,rinv20);
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq20             = _mm_mul_pd(iq2,jq0);
 374
 375             /* EWALD ELECTROSTATICS */
 376
 377             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 378             ewrt             = _mm_mul_pd(r20,ewtabscale);
 379             ewitab           = _mm_cvttpd_epi32(ewrt);
 380 #ifdef __XOP__
 381             eweps            = _mm_frcz_pd(ewrt);
 382 #else
 383             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 384 #endif
 385             twoeweps         = _mm_add_pd(eweps,eweps);
 386             ewitab           = _mm_slli_epi32(ewitab,2);
 387             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 388             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 389             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 390             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 391             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 392             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 393             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 394             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 395             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 396             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 397
 398             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velec            = _mm_and_pd(velec,cutoff_mask);
 402             velecsum         = _mm_add_pd(velecsum,velec);
 403
 404             fscal            = felec;
 405
 406             fscal            = _mm_and_pd(fscal,cutoff_mask);
 407
 408             /* Update vectorial force */
 409             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 410             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 411             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 412
 413             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 414             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 415             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 416
 417             }
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             if (gmx_mm_any_lt(rsq30,rcutoff2))
 424             {
 425
 426             r30              = _mm_mul_pd(rsq30,rinv30);
 427
 428             /* Compute parameters for interactions between i and j atoms */
 429             qq30             = _mm_mul_pd(iq3,jq0);
 430
 431             /* EWALD ELECTROSTATICS */
 432
 433             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 434             ewrt             = _mm_mul_pd(r30,ewtabscale);
 435             ewitab           = _mm_cvttpd_epi32(ewrt);
 436 #ifdef __XOP__
 437             eweps            = _mm_frcz_pd(ewrt);
 438 #else
 439             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 440 #endif
 441             twoeweps         = _mm_add_pd(eweps,eweps);
 442             ewitab           = _mm_slli_epi32(ewitab,2);
 443             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 444             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 445             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 446             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 447             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 448             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 449             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 450             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 451             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 452             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 453
 454             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velec            = _mm_and_pd(velec,cutoff_mask);
 458             velecsum         = _mm_add_pd(velecsum,velec);
 459
 460             fscal            = felec;
 461
 462             fscal            = _mm_and_pd(fscal,cutoff_mask);
 463
 464             /* Update vectorial force */
 465             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 466             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 467             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 468
 469             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 470             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 471             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 472
 473             }
 474
 475             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 476
 477             /* Inner loop uses 208 flops */
 478         }
 479
 480         if(jidx<j_index_end)
 481         {
 482
 483             jnrA             = jjnr[jidx];
 484             j_coord_offsetA  = DIM*jnrA;
 485
 486             /* load j atom coordinates */
 487             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 488                                               &jx0,&jy0,&jz0);
 489
 490             /* Calculate displacement vector */
 491             dx00             = _mm_sub_pd(ix0,jx0);
 492             dy00             = _mm_sub_pd(iy0,jy0);
 493             dz00             = _mm_sub_pd(iz0,jz0);
 494             dx10             = _mm_sub_pd(ix1,jx0);
 495             dy10             = _mm_sub_pd(iy1,jy0);
 496             dz10             = _mm_sub_pd(iz1,jz0);
 497             dx20             = _mm_sub_pd(ix2,jx0);
 498             dy20             = _mm_sub_pd(iy2,jy0);
 499             dz20             = _mm_sub_pd(iz2,jz0);
 500             dx30             = _mm_sub_pd(ix3,jx0);
 501             dy30             = _mm_sub_pd(iy3,jy0);
 502             dz30             = _mm_sub_pd(iz3,jz0);
 503
 504             /* Calculate squared distance and things based on it */
 505             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 506             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 507             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 508             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 509
 510             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 511             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 512             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 513             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 514
 515             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 516             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 517             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 518             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 519
 520             /* Load parameters for j particles */
 521             jq0              = _mm_load_sd(charge+jnrA+0);
 522             vdwjidx0A        = 2*vdwtype[jnrA+0];
 523
 524             fjx0             = _mm_setzero_pd();
 525             fjy0             = _mm_setzero_pd();
 526             fjz0             = _mm_setzero_pd();
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             if (gmx_mm_any_lt(rsq00,rcutoff2))
 533             {
 534
 535             r00              = _mm_mul_pd(rsq00,rinv00);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 539             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 540
 541             /* Analytical LJ-PME */
 542             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 543             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 544             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 545             exponent         = gmx_simd_exp_d(ewcljrsq);
 546             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 547             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 548             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 549             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 550             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 551             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 552                                _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));
 553             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 554             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);
 555
 556             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 560             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 561             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 562
 563             fscal            = fvdw;
 564
 565             fscal            = _mm_and_pd(fscal,cutoff_mask);
 566
 567             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 568
 569             /* Update vectorial force */
 570             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 571             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 572             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 573
 574             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 575             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 576             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 577
 578             }
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             if (gmx_mm_any_lt(rsq10,rcutoff2))
 585             {
 586
 587             r10              = _mm_mul_pd(rsq10,rinv10);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq10             = _mm_mul_pd(iq1,jq0);
 591
 592             /* EWALD ELECTROSTATICS */
 593
 594             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 595             ewrt             = _mm_mul_pd(r10,ewtabscale);
 596             ewitab           = _mm_cvttpd_epi32(ewrt);
 597 #ifdef __XOP__
 598             eweps            = _mm_frcz_pd(ewrt);
 599 #else
 600             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 601 #endif
 602             twoeweps         = _mm_add_pd(eweps,eweps);
 603             ewitab           = _mm_slli_epi32(ewitab,2);
 604             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 605             ewtabD           = _mm_setzero_pd();
 606             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 607             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 608             ewtabFn          = _mm_setzero_pd();
 609             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 610             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 611             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 612             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 613             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 614
 615             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 616
 617             /* Update potential sum for this i atom from the interaction with this j atom. */
 618             velec            = _mm_and_pd(velec,cutoff_mask);
 619             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 620             velecsum         = _mm_add_pd(velecsum,velec);
 621
 622             fscal            = felec;
 623
 624             fscal            = _mm_and_pd(fscal,cutoff_mask);
 625
 626             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 627
 628             /* Update vectorial force */
 629             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 630             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 631             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 632
 633             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 634             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 635             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 636
 637             }
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             if (gmx_mm_any_lt(rsq20,rcutoff2))
 644             {
 645
 646             r20              = _mm_mul_pd(rsq20,rinv20);
 647
 648             /* Compute parameters for interactions between i and j atoms */
 649             qq20             = _mm_mul_pd(iq2,jq0);
 650
 651             /* EWALD ELECTROSTATICS */
 652
 653             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 654             ewrt             = _mm_mul_pd(r20,ewtabscale);
 655             ewitab           = _mm_cvttpd_epi32(ewrt);
 656 #ifdef __XOP__
 657             eweps            = _mm_frcz_pd(ewrt);
 658 #else
 659             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 660 #endif
 661             twoeweps         = _mm_add_pd(eweps,eweps);
 662             ewitab           = _mm_slli_epi32(ewitab,2);
 663             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 664             ewtabD           = _mm_setzero_pd();
 665             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 666             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 667             ewtabFn          = _mm_setzero_pd();
 668             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 669             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 670             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 671             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 672             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 673
 674             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 675
 676             /* Update potential sum for this i atom from the interaction with this j atom. */
 677             velec            = _mm_and_pd(velec,cutoff_mask);
 678             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 679             velecsum         = _mm_add_pd(velecsum,velec);
 680
 681             fscal            = felec;
 682
 683             fscal            = _mm_and_pd(fscal,cutoff_mask);
 684
 685             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 686
 687             /* Update vectorial force */
 688             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 689             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 690             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 691
 692             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 693             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 694             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 695
 696             }
 697
 698             /**************************
 699              * CALCULATE INTERACTIONS *
 700              **************************/
 701
 702             if (gmx_mm_any_lt(rsq30,rcutoff2))
 703             {
 704
 705             r30              = _mm_mul_pd(rsq30,rinv30);
 706
 707             /* Compute parameters for interactions between i and j atoms */
 708             qq30             = _mm_mul_pd(iq3,jq0);
 709
 710             /* EWALD ELECTROSTATICS */
 711
 712             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 713             ewrt             = _mm_mul_pd(r30,ewtabscale);
 714             ewitab           = _mm_cvttpd_epi32(ewrt);
 715 #ifdef __XOP__
 716             eweps            = _mm_frcz_pd(ewrt);
 717 #else
 718             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 719 #endif
 720             twoeweps         = _mm_add_pd(eweps,eweps);
 721             ewitab           = _mm_slli_epi32(ewitab,2);
 722             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 723             ewtabD           = _mm_setzero_pd();
 724             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 725             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 726             ewtabFn          = _mm_setzero_pd();
 727             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 728             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 729             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 730             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 731             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 732
 733             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 734
 735             /* Update potential sum for this i atom from the interaction with this j atom. */
 736             velec            = _mm_and_pd(velec,cutoff_mask);
 737             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 738             velecsum         = _mm_add_pd(velecsum,velec);
 739
 740             fscal            = felec;
 741
 742             fscal            = _mm_and_pd(fscal,cutoff_mask);
 743
 744             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 745
 746             /* Update vectorial force */
 747             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 748             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 749             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 750
 751             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 752             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 753             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 754
 755             }
 756
 757             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 758
 759             /* Inner loop uses 208 flops */
 760         }
 761
 762         /* End of innermost loop */
 763
 764         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 765                                               f+i_coord_offset,fshift+i_shift_offset);
 766
 767         ggid                        = gid[iidx];
 768         /* Update potential energies */
 769         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 770         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 771
 772         /* Increment number of inner iterations */
 773         inneriter                  += j_index_end - j_index_start;
 774
 775         /* Outer loop uses 26 flops */
 776     }
 777
 778     /* Increment number of outer iterations */
 779     outeriter        += nri;
 780
 781     /* Update outer/inner flops */
 782
 783     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*208);
 784 }
 785 /*
 786  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_128_fma_double
 787  * Electrostatics interaction: Ewald
 788  * VdW interaction:            LJEwald
 789  * Geometry:                   Water4-Particle
 790  * Calculate force/pot:        Force
 791  */
 792 void
 793 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_128_fma_double
 794                     (t_nblist                    * gmx_restrict       nlist,
 795                      rvec                        * gmx_restrict          xx,
 796                      rvec                        * gmx_restrict          ff,
 797                      t_forcerec                  * gmx_restrict          fr,
 798                      t_mdatoms                   * gmx_restrict     mdatoms,
 799                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 800                      t_nrnb                      * gmx_restrict        nrnb)
 801 {
 802     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 803      * just 0 for non-waters.
 804      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 805      * jnr indices corresponding to data put in the four positions in the SIMD register.
 806      */
 807     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 808     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 809     int              jnrA,jnrB;
 810     int              j_coord_offsetA,j_coord_offsetB;
 811     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 812     real             rcutoff_scalar;
 813     real             *shiftvec,*fshift,*x,*f;
 814     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 815     int              vdwioffset0;
 816     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 817     int              vdwioffset1;
 818     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 819     int              vdwioffset2;
 820     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 821     int              vdwioffset3;
 822     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 823     int              vdwjidx0A,vdwjidx0B;
 824     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 825     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 826     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 827     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 828     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 829     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 830     real             *charge;
 831     int              nvdwtype;
 832     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 833     int              *vdwtype;
 834     real             *vdwparam;
 835     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 836     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 837     __m128d           c6grid_00;
 838     __m128d           c6grid_10;
 839     __m128d           c6grid_20;
 840     __m128d           c6grid_30;
 841     real             *vdwgridparam;
 842     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 843     __m128d           one_half  = _mm_set1_pd(0.5);
 844     __m128d           minus_one = _mm_set1_pd(-1.0);
 845     __m128i          ewitab;
 846     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 847     real             *ewtab;
 848     __m128d          dummy_mask,cutoff_mask;
 849     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 850     __m128d          one     = _mm_set1_pd(1.0);
 851     __m128d          two     = _mm_set1_pd(2.0);
 852     x                = xx[0];
 853     f                = ff[0];
 854
 855     nri              = nlist->nri;
 856     iinr             = nlist->iinr;
 857     jindex           = nlist->jindex;
 858     jjnr             = nlist->jjnr;
 859     shiftidx         = nlist->shift;
 860     gid              = nlist->gid;
 861     shiftvec         = fr->shift_vec[0];
 862     fshift           = fr->fshift[0];
 863     facel            = _mm_set1_pd(fr->epsfac);
 864     charge           = mdatoms->chargeA;
 865     nvdwtype         = fr->ntype;
 866     vdwparam         = fr->nbfp;
 867     vdwtype          = mdatoms->typeA;
 868     vdwgridparam     = fr->ljpme_c6grid;
 869     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 870     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 871     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 872
 873     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 874     ewtab            = fr->ic->tabq_coul_F;
 875     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 876     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 877
 878     /* Setup water-specific parameters */
 879     inr              = nlist->iinr[0];
 880     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 881     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 882     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 883     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 884
 885     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 886     rcutoff_scalar   = fr->rcoulomb;
 887     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 888     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 889
 890     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 891     rvdw             = _mm_set1_pd(fr->rvdw);
 892
 893     /* Avoid stupid compiler warnings */
 894     jnrA = jnrB = 0;
 895     j_coord_offsetA = 0;
 896     j_coord_offsetB = 0;
 897
 898     outeriter        = 0;
 899     inneriter        = 0;
 900
 901     /* Start outer loop over neighborlists */
 902     for(iidx=0; iidx<nri; iidx++)
 903     {
 904         /* Load shift vector for this list */
 905         i_shift_offset   = DIM*shiftidx[iidx];
 906
 907         /* Load limits for loop over neighbors */
 908         j_index_start    = jindex[iidx];
 909         j_index_end      = jindex[iidx+1];
 910
 911         /* Get outer coordinate index */
 912         inr              = iinr[iidx];
 913         i_coord_offset   = DIM*inr;
 914
 915         /* Load i particle coords and add shift vector */
 916         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 917                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 918
 919         fix0             = _mm_setzero_pd();
 920         fiy0             = _mm_setzero_pd();
 921         fiz0             = _mm_setzero_pd();
 922         fix1             = _mm_setzero_pd();
 923         fiy1             = _mm_setzero_pd();
 924         fiz1             = _mm_setzero_pd();
 925         fix2             = _mm_setzero_pd();
 926         fiy2             = _mm_setzero_pd();
 927         fiz2             = _mm_setzero_pd();
 928         fix3             = _mm_setzero_pd();
 929         fiy3             = _mm_setzero_pd();
 930         fiz3             = _mm_setzero_pd();
 931
 932         /* Start inner kernel loop */
 933         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 934         {
 935
 936             /* Get j neighbor index, and coordinate index */
 937             jnrA             = jjnr[jidx];
 938             jnrB             = jjnr[jidx+1];
 939             j_coord_offsetA  = DIM*jnrA;
 940             j_coord_offsetB  = DIM*jnrB;
 941
 942             /* load j atom coordinates */
 943             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 944                                               &jx0,&jy0,&jz0);
 945
 946             /* Calculate displacement vector */
 947             dx00             = _mm_sub_pd(ix0,jx0);
 948             dy00             = _mm_sub_pd(iy0,jy0);
 949             dz00             = _mm_sub_pd(iz0,jz0);
 950             dx10             = _mm_sub_pd(ix1,jx0);
 951             dy10             = _mm_sub_pd(iy1,jy0);
 952             dz10             = _mm_sub_pd(iz1,jz0);
 953             dx20             = _mm_sub_pd(ix2,jx0);
 954             dy20             = _mm_sub_pd(iy2,jy0);
 955             dz20             = _mm_sub_pd(iz2,jz0);
 956             dx30             = _mm_sub_pd(ix3,jx0);
 957             dy30             = _mm_sub_pd(iy3,jy0);
 958             dz30             = _mm_sub_pd(iz3,jz0);
 959
 960             /* Calculate squared distance and things based on it */
 961             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 962             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 963             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 964             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 965
 966             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 967             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 968             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 969             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 970
 971             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 972             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 973             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 974             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 975
 976             /* Load parameters for j particles */
 977             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 978             vdwjidx0A        = 2*vdwtype[jnrA+0];
 979             vdwjidx0B        = 2*vdwtype[jnrB+0];
 980
 981             fjx0             = _mm_setzero_pd();
 982             fjy0             = _mm_setzero_pd();
 983             fjz0             = _mm_setzero_pd();
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             if (gmx_mm_any_lt(rsq00,rcutoff2))
 990             {
 991
 992             r00              = _mm_mul_pd(rsq00,rinv00);
 993
 994             /* Compute parameters for interactions between i and j atoms */
 995             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 996                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 997             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 998                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 999
1000             /* Analytical LJ-PME */
1001             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1002             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1003             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1004             exponent         = gmx_simd_exp_d(ewcljrsq);
1005             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1006             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1007             /* f6A = 6 * C6grid * (1 - poly) */
1008             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1009             /* f6B = C6grid * exponent * beta^6 */
1010             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1011             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1012             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1013
1014             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1015
1016             fscal            = fvdw;
1017
1018             fscal            = _mm_and_pd(fscal,cutoff_mask);
1019
1020             /* Update vectorial force */
1021             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1022             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1023             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1024
1025             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1026             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1027             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1028
1029             }
1030
1031             /**************************
1032              * CALCULATE INTERACTIONS *
1033              **************************/
1034
1035             if (gmx_mm_any_lt(rsq10,rcutoff2))
1036             {
1037
1038             r10              = _mm_mul_pd(rsq10,rinv10);
1039
1040             /* Compute parameters for interactions between i and j atoms */
1041             qq10             = _mm_mul_pd(iq1,jq0);
1042
1043             /* EWALD ELECTROSTATICS */
1044
1045             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1046             ewrt             = _mm_mul_pd(r10,ewtabscale);
1047             ewitab           = _mm_cvttpd_epi32(ewrt);
1048 #ifdef __XOP__
1049             eweps            = _mm_frcz_pd(ewrt);
1050 #else
1051             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1052 #endif
1053             twoeweps         = _mm_add_pd(eweps,eweps);
1054             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1055                                          &ewtabF,&ewtabFn);
1056             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1057             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1058
1059             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1060
1061             fscal            = felec;
1062
1063             fscal            = _mm_and_pd(fscal,cutoff_mask);
1064
1065             /* Update vectorial force */
1066             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1067             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1068             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1069
1070             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1071             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1072             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1073
1074             }
1075
1076             /**************************
1077              * CALCULATE INTERACTIONS *
1078              **************************/
1079
1080             if (gmx_mm_any_lt(rsq20,rcutoff2))
1081             {
1082
1083             r20              = _mm_mul_pd(rsq20,rinv20);
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             qq20             = _mm_mul_pd(iq2,jq0);
1087
1088             /* EWALD ELECTROSTATICS */
1089
1090             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1091             ewrt             = _mm_mul_pd(r20,ewtabscale);
1092             ewitab           = _mm_cvttpd_epi32(ewrt);
1093 #ifdef __XOP__
1094             eweps            = _mm_frcz_pd(ewrt);
1095 #else
1096             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1097 #endif
1098             twoeweps         = _mm_add_pd(eweps,eweps);
1099             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1100                                          &ewtabF,&ewtabFn);
1101             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1102             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1103
1104             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1105
1106             fscal            = felec;
1107
1108             fscal            = _mm_and_pd(fscal,cutoff_mask);
1109
1110             /* Update vectorial force */
1111             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1112             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1113             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1114
1115             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1116             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1117             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1118
1119             }
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             if (gmx_mm_any_lt(rsq30,rcutoff2))
1126             {
1127
1128             r30              = _mm_mul_pd(rsq30,rinv30);
1129
1130             /* Compute parameters for interactions between i and j atoms */
1131             qq30             = _mm_mul_pd(iq3,jq0);
1132
1133             /* EWALD ELECTROSTATICS */
1134
1135             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1136             ewrt             = _mm_mul_pd(r30,ewtabscale);
1137             ewitab           = _mm_cvttpd_epi32(ewrt);
1138 #ifdef __XOP__
1139             eweps            = _mm_frcz_pd(ewrt);
1140 #else
1141             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1142 #endif
1143             twoeweps         = _mm_add_pd(eweps,eweps);
1144             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1145                                          &ewtabF,&ewtabFn);
1146             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1147             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1148
1149             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1150
1151             fscal            = felec;
1152
1153             fscal            = _mm_and_pd(fscal,cutoff_mask);
1154
1155             /* Update vectorial force */
1156             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1157             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1158             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1159
1160             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1161             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1162             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1163
1164             }
1165
1166             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1167
1168             /* Inner loop uses 179 flops */
1169         }
1170
1171         if(jidx<j_index_end)
1172         {
1173
1174             jnrA             = jjnr[jidx];
1175             j_coord_offsetA  = DIM*jnrA;
1176
1177             /* load j atom coordinates */
1178             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1179                                               &jx0,&jy0,&jz0);
1180
1181             /* Calculate displacement vector */
1182             dx00             = _mm_sub_pd(ix0,jx0);
1183             dy00             = _mm_sub_pd(iy0,jy0);
1184             dz00             = _mm_sub_pd(iz0,jz0);
1185             dx10             = _mm_sub_pd(ix1,jx0);
1186             dy10             = _mm_sub_pd(iy1,jy0);
1187             dz10             = _mm_sub_pd(iz1,jz0);
1188             dx20             = _mm_sub_pd(ix2,jx0);
1189             dy20             = _mm_sub_pd(iy2,jy0);
1190             dz20             = _mm_sub_pd(iz2,jz0);
1191             dx30             = _mm_sub_pd(ix3,jx0);
1192             dy30             = _mm_sub_pd(iy3,jy0);
1193             dz30             = _mm_sub_pd(iz3,jz0);
1194
1195             /* Calculate squared distance and things based on it */
1196             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1197             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1198             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1199             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1200
1201             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1202             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1203             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1204             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1205
1206             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1207             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1208             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1209             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1210
1211             /* Load parameters for j particles */
1212             jq0              = _mm_load_sd(charge+jnrA+0);
1213             vdwjidx0A        = 2*vdwtype[jnrA+0];
1214
1215             fjx0             = _mm_setzero_pd();
1216             fjy0             = _mm_setzero_pd();
1217             fjz0             = _mm_setzero_pd();
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             if (gmx_mm_any_lt(rsq00,rcutoff2))
1224             {
1225
1226             r00              = _mm_mul_pd(rsq00,rinv00);
1227
1228             /* Compute parameters for interactions between i and j atoms */
1229             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1230             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1231
1232             /* Analytical LJ-PME */
1233             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1234             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1235             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1236             exponent         = gmx_simd_exp_d(ewcljrsq);
1237             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1238             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1239             /* f6A = 6 * C6grid * (1 - poly) */
1240             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1241             /* f6B = C6grid * exponent * beta^6 */
1242             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1243             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1244             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1245
1246             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1247
1248             fscal            = fvdw;
1249
1250             fscal            = _mm_and_pd(fscal,cutoff_mask);
1251
1252             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1253
1254             /* Update vectorial force */
1255             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1256             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1257             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1258
1259             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1260             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1261             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1262
1263             }
1264
1265             /**************************
1266              * CALCULATE INTERACTIONS *
1267              **************************/
1268
1269             if (gmx_mm_any_lt(rsq10,rcutoff2))
1270             {
1271
1272             r10              = _mm_mul_pd(rsq10,rinv10);
1273
1274             /* Compute parameters for interactions between i and j atoms */
1275             qq10             = _mm_mul_pd(iq1,jq0);
1276
1277             /* EWALD ELECTROSTATICS */
1278
1279             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1280             ewrt             = _mm_mul_pd(r10,ewtabscale);
1281             ewitab           = _mm_cvttpd_epi32(ewrt);
1282 #ifdef __XOP__
1283             eweps            = _mm_frcz_pd(ewrt);
1284 #else
1285             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1286 #endif
1287             twoeweps         = _mm_add_pd(eweps,eweps);
1288             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1289             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1290             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1291
1292             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1293
1294             fscal            = felec;
1295
1296             fscal            = _mm_and_pd(fscal,cutoff_mask);
1297
1298             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1299
1300             /* Update vectorial force */
1301             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1302             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1303             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1304
1305             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1306             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1307             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1308
1309             }
1310
1311             /**************************
1312              * CALCULATE INTERACTIONS *
1313              **************************/
1314
1315             if (gmx_mm_any_lt(rsq20,rcutoff2))
1316             {
1317
1318             r20              = _mm_mul_pd(rsq20,rinv20);
1319
1320             /* Compute parameters for interactions between i and j atoms */
1321             qq20             = _mm_mul_pd(iq2,jq0);
1322
1323             /* EWALD ELECTROSTATICS */
1324
1325             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1326             ewrt             = _mm_mul_pd(r20,ewtabscale);
1327             ewitab           = _mm_cvttpd_epi32(ewrt);
1328 #ifdef __XOP__
1329             eweps            = _mm_frcz_pd(ewrt);
1330 #else
1331             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1332 #endif
1333             twoeweps         = _mm_add_pd(eweps,eweps);
1334             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1335             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1336             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1337
1338             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1339
1340             fscal            = felec;
1341
1342             fscal            = _mm_and_pd(fscal,cutoff_mask);
1343
1344             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1345
1346             /* Update vectorial force */
1347             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1348             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1349             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1350
1351             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1352             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1353             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1354
1355             }
1356
1357             /**************************
1358              * CALCULATE INTERACTIONS *
1359              **************************/
1360
1361             if (gmx_mm_any_lt(rsq30,rcutoff2))
1362             {
1363
1364             r30              = _mm_mul_pd(rsq30,rinv30);
1365
1366             /* Compute parameters for interactions between i and j atoms */
1367             qq30             = _mm_mul_pd(iq3,jq0);
1368
1369             /* EWALD ELECTROSTATICS */
1370
1371             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1372             ewrt             = _mm_mul_pd(r30,ewtabscale);
1373             ewitab           = _mm_cvttpd_epi32(ewrt);
1374 #ifdef __XOP__
1375             eweps            = _mm_frcz_pd(ewrt);
1376 #else
1377             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1378 #endif
1379             twoeweps         = _mm_add_pd(eweps,eweps);
1380             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1381             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1382             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1383
1384             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1385
1386             fscal            = felec;
1387
1388             fscal            = _mm_and_pd(fscal,cutoff_mask);
1389
1390             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1391
1392             /* Update vectorial force */
1393             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1394             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1395             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1396
1397             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1398             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1399             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1400
1401             }
1402
1403             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1404
1405             /* Inner loop uses 179 flops */
1406         }
1407
1408         /* End of innermost loop */
1409
1410         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1411                                               f+i_coord_offset,fshift+i_shift_offset);
1412
1413         /* Increment number of inner iterations */
1414         inneriter                  += j_index_end - j_index_start;
1415
1416         /* Outer loop uses 24 flops */
1417     }
1418
1419     /* Increment number of outer iterations */
1420     outeriter        += nri;
1421
1422     /* Update outer/inner flops */
1423
1424     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*179);
1425 }