src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     int              vdwjidx1A,vdwjidx1B;
  89     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  90     int              vdwjidx2A,vdwjidx2B;
  91     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  94     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  95     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  97     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  98     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 100     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 101     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 102     real             *charge;
 103     int              nvdwtype;
 104     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 105     int              *vdwtype;
 106     real             *vdwparam;
 107     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 108     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 109     __m128d           c6grid_00;
 110     __m128d           c6grid_01;
 111     __m128d           c6grid_02;
 112     __m128d           c6grid_10;
 113     __m128d           c6grid_11;
 114     __m128d           c6grid_12;
 115     __m128d           c6grid_20;
 116     __m128d           c6grid_21;
 117     __m128d           c6grid_22;
 118     real             *vdwgridparam;
 119     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 120     __m128d           one_half  = _mm_set1_pd(0.5);
 121     __m128d           minus_one = _mm_set1_pd(-1.0);
 122     __m128i          ewitab;
 123     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 124     real             *ewtab;
 125     __m128d          dummy_mask,cutoff_mask;
 126     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 127     __m128d          one     = _mm_set1_pd(1.0);
 128     __m128d          two     = _mm_set1_pd(2.0);
 129     x                = xx[0];
 130     f                = ff[0];
 131
 132     nri              = nlist->nri;
 133     iinr             = nlist->iinr;
 134     jindex           = nlist->jindex;
 135     jjnr             = nlist->jjnr;
 136     shiftidx         = nlist->shift;
 137     gid              = nlist->gid;
 138     shiftvec         = fr->shift_vec[0];
 139     fshift           = fr->fshift[0];
 140     facel            = _mm_set1_pd(fr->epsfac);
 141     charge           = mdatoms->chargeA;
 142     nvdwtype         = fr->ntype;
 143     vdwparam         = fr->nbfp;
 144     vdwtype          = mdatoms->typeA;
 145     vdwgridparam     = fr->ljpme_c6grid;
 146     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 147     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 148     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 149
 150     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 151     ewtab            = fr->ic->tabq_coul_FDV0;
 152     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 153     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 154
 155     /* Setup water-specific parameters */
 156     inr              = nlist->iinr[0];
 157     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 158     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 159     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 160     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162     jq0              = _mm_set1_pd(charge[inr+0]);
 163     jq1              = _mm_set1_pd(charge[inr+1]);
 164     jq2              = _mm_set1_pd(charge[inr+2]);
 165     vdwjidx0A        = 2*vdwtype[inr+0];
 166     qq00             = _mm_mul_pd(iq0,jq0);
 167     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 168     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 169     c6grid_00        = _mm_set1_pd(vdwgridparam[vdwioffset0+vdwjidx0A]);
 170     qq01             = _mm_mul_pd(iq0,jq1);
 171     qq02             = _mm_mul_pd(iq0,jq2);
 172     qq10             = _mm_mul_pd(iq1,jq0);
 173     qq11             = _mm_mul_pd(iq1,jq1);
 174     qq12             = _mm_mul_pd(iq1,jq2);
 175     qq20             = _mm_mul_pd(iq2,jq0);
 176     qq21             = _mm_mul_pd(iq2,jq1);
 177     qq22             = _mm_mul_pd(iq2,jq2);
 178
 179     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 180     rcutoff_scalar   = fr->rcoulomb;
 181     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 182     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 183
 184     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 185     rvdw             = _mm_set1_pd(fr->rvdw);
 186
 187     /* Avoid stupid compiler warnings */
 188     jnrA = jnrB = 0;
 189     j_coord_offsetA = 0;
 190     j_coord_offsetB = 0;
 191
 192     outeriter        = 0;
 193     inneriter        = 0;
 194
 195     /* Start outer loop over neighborlists */
 196     for(iidx=0; iidx<nri; iidx++)
 197     {
 198         /* Load shift vector for this list */
 199         i_shift_offset   = DIM*shiftidx[iidx];
 200
 201         /* Load limits for loop over neighbors */
 202         j_index_start    = jindex[iidx];
 203         j_index_end      = jindex[iidx+1];
 204
 205         /* Get outer coordinate index */
 206         inr              = iinr[iidx];
 207         i_coord_offset   = DIM*inr;
 208
 209         /* Load i particle coords and add shift vector */
 210         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 211                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 212
 213         fix0             = _mm_setzero_pd();
 214         fiy0             = _mm_setzero_pd();
 215         fiz0             = _mm_setzero_pd();
 216         fix1             = _mm_setzero_pd();
 217         fiy1             = _mm_setzero_pd();
 218         fiz1             = _mm_setzero_pd();
 219         fix2             = _mm_setzero_pd();
 220         fiy2             = _mm_setzero_pd();
 221         fiz2             = _mm_setzero_pd();
 222
 223         /* Reset potential sums */
 224         velecsum         = _mm_setzero_pd();
 225         vvdwsum          = _mm_setzero_pd();
 226
 227         /* Start inner kernel loop */
 228         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 229         {
 230
 231             /* Get j neighbor index, and coordinate index */
 232             jnrA             = jjnr[jidx];
 233             jnrB             = jjnr[jidx+1];
 234             j_coord_offsetA  = DIM*jnrA;
 235             j_coord_offsetB  = DIM*jnrB;
 236
 237             /* load j atom coordinates */
 238             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 239                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 240
 241             /* Calculate displacement vector */
 242             dx00             = _mm_sub_pd(ix0,jx0);
 243             dy00             = _mm_sub_pd(iy0,jy0);
 244             dz00             = _mm_sub_pd(iz0,jz0);
 245             dx01             = _mm_sub_pd(ix0,jx1);
 246             dy01             = _mm_sub_pd(iy0,jy1);
 247             dz01             = _mm_sub_pd(iz0,jz1);
 248             dx02             = _mm_sub_pd(ix0,jx2);
 249             dy02             = _mm_sub_pd(iy0,jy2);
 250             dz02             = _mm_sub_pd(iz0,jz2);
 251             dx10             = _mm_sub_pd(ix1,jx0);
 252             dy10             = _mm_sub_pd(iy1,jy0);
 253             dz10             = _mm_sub_pd(iz1,jz0);
 254             dx11             = _mm_sub_pd(ix1,jx1);
 255             dy11             = _mm_sub_pd(iy1,jy1);
 256             dz11             = _mm_sub_pd(iz1,jz1);
 257             dx12             = _mm_sub_pd(ix1,jx2);
 258             dy12             = _mm_sub_pd(iy1,jy2);
 259             dz12             = _mm_sub_pd(iz1,jz2);
 260             dx20             = _mm_sub_pd(ix2,jx0);
 261             dy20             = _mm_sub_pd(iy2,jy0);
 262             dz20             = _mm_sub_pd(iz2,jz0);
 263             dx21             = _mm_sub_pd(ix2,jx1);
 264             dy21             = _mm_sub_pd(iy2,jy1);
 265             dz21             = _mm_sub_pd(iz2,jz1);
 266             dx22             = _mm_sub_pd(ix2,jx2);
 267             dy22             = _mm_sub_pd(iy2,jy2);
 268             dz22             = _mm_sub_pd(iz2,jz2);
 269
 270             /* Calculate squared distance and things based on it */
 271             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 272             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 273             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 274             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 275             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 276             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 277             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 278             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 279             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 280
 281             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 282             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 283             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 284             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 285             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 286             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 287             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 288             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 289             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 290
 291             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 292             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 293             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 294             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 295             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 296             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 297             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 298             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 299             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 300
 301             fjx0             = _mm_setzero_pd();
 302             fjy0             = _mm_setzero_pd();
 303             fjz0             = _mm_setzero_pd();
 304             fjx1             = _mm_setzero_pd();
 305             fjy1             = _mm_setzero_pd();
 306             fjz1             = _mm_setzero_pd();
 307             fjx2             = _mm_setzero_pd();
 308             fjy2             = _mm_setzero_pd();
 309             fjz2             = _mm_setzero_pd();
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             if (gmx_mm_any_lt(rsq00,rcutoff2))
 316             {
 317
 318             r00              = _mm_mul_pd(rsq00,rinv00);
 319
 320             /* EWALD ELECTROSTATICS */
 321
 322             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 323             ewrt             = _mm_mul_pd(r00,ewtabscale);
 324             ewitab           = _mm_cvttpd_epi32(ewrt);
 325 #ifdef __XOP__
 326             eweps            = _mm_frcz_pd(ewrt);
 327 #else
 328             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 329 #endif
 330             twoeweps         = _mm_add_pd(eweps,eweps);
 331             ewitab           = _mm_slli_epi32(ewitab,2);
 332             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 333             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 334             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 335             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 336             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 337             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 338             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 339             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 340             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 341             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 342
 343             /* Analytical LJ-PME */
 344             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 345             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 346             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 347             exponent         = gmx_simd_exp_d(ewcljrsq);
 348             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 349             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 350             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 351             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 352             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 353             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 354                                _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));
 355             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 356             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);
 357
 358             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velec            = _mm_and_pd(velec,cutoff_mask);
 362             velecsum         = _mm_add_pd(velecsum,velec);
 363             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 364             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 365
 366             fscal            = _mm_add_pd(felec,fvdw);
 367
 368             fscal            = _mm_and_pd(fscal,cutoff_mask);
 369
 370             /* Update vectorial force */
 371             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 372             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 373             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 374
 375             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 376             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 377             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 378
 379             }
 380
 381             /**************************
 382              * CALCULATE INTERACTIONS *
 383              **************************/
 384
 385             if (gmx_mm_any_lt(rsq01,rcutoff2))
 386             {
 387
 388             r01              = _mm_mul_pd(rsq01,rinv01);
 389
 390             /* EWALD ELECTROSTATICS */
 391
 392             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 393             ewrt             = _mm_mul_pd(r01,ewtabscale);
 394             ewitab           = _mm_cvttpd_epi32(ewrt);
 395 #ifdef __XOP__
 396             eweps            = _mm_frcz_pd(ewrt);
 397 #else
 398             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 399 #endif
 400             twoeweps         = _mm_add_pd(eweps,eweps);
 401             ewitab           = _mm_slli_epi32(ewitab,2);
 402             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 403             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 404             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 405             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 406             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 407             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 408             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 409             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 410             velec            = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
 411             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 412
 413             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
 414
 415             /* Update potential sum for this i atom from the interaction with this j atom. */
 416             velec            = _mm_and_pd(velec,cutoff_mask);
 417             velecsum         = _mm_add_pd(velecsum,velec);
 418
 419             fscal            = felec;
 420
 421             fscal            = _mm_and_pd(fscal,cutoff_mask);
 422
 423             /* Update vectorial force */
 424             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 425             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 426             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 427
 428             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 429             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 430             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 431
 432             }
 433
 434             /**************************
 435              * CALCULATE INTERACTIONS *
 436              **************************/
 437
 438             if (gmx_mm_any_lt(rsq02,rcutoff2))
 439             {
 440
 441             r02              = _mm_mul_pd(rsq02,rinv02);
 442
 443             /* EWALD ELECTROSTATICS */
 444
 445             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 446             ewrt             = _mm_mul_pd(r02,ewtabscale);
 447             ewitab           = _mm_cvttpd_epi32(ewrt);
 448 #ifdef __XOP__
 449             eweps            = _mm_frcz_pd(ewrt);
 450 #else
 451             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 452 #endif
 453             twoeweps         = _mm_add_pd(eweps,eweps);
 454             ewitab           = _mm_slli_epi32(ewitab,2);
 455             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 456             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 457             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 458             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 459             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 460             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 461             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 462             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 463             velec            = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
 464             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 465
 466             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
 467
 468             /* Update potential sum for this i atom from the interaction with this j atom. */
 469             velec            = _mm_and_pd(velec,cutoff_mask);
 470             velecsum         = _mm_add_pd(velecsum,velec);
 471
 472             fscal            = felec;
 473
 474             fscal            = _mm_and_pd(fscal,cutoff_mask);
 475
 476             /* Update vectorial force */
 477             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 478             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 479             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 480
 481             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 482             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 483             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 484
 485             }
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             if (gmx_mm_any_lt(rsq10,rcutoff2))
 492             {
 493
 494             r10              = _mm_mul_pd(rsq10,rinv10);
 495
 496             /* EWALD ELECTROSTATICS */
 497
 498             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 499             ewrt             = _mm_mul_pd(r10,ewtabscale);
 500             ewitab           = _mm_cvttpd_epi32(ewrt);
 501 #ifdef __XOP__
 502             eweps            = _mm_frcz_pd(ewrt);
 503 #else
 504             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 505 #endif
 506             twoeweps         = _mm_add_pd(eweps,eweps);
 507             ewitab           = _mm_slli_epi32(ewitab,2);
 508             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 509             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 510             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 511             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 512             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 513             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 514             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 515             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 516             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 517             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 518
 519             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 520
 521             /* Update potential sum for this i atom from the interaction with this j atom. */
 522             velec            = _mm_and_pd(velec,cutoff_mask);
 523             velecsum         = _mm_add_pd(velecsum,velec);
 524
 525             fscal            = felec;
 526
 527             fscal            = _mm_and_pd(fscal,cutoff_mask);
 528
 529             /* Update vectorial force */
 530             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 531             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 532             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 533
 534             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 535             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 536             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 537
 538             }
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             if (gmx_mm_any_lt(rsq11,rcutoff2))
 545             {
 546
 547             r11              = _mm_mul_pd(rsq11,rinv11);
 548
 549             /* EWALD ELECTROSTATICS */
 550
 551             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 552             ewrt             = _mm_mul_pd(r11,ewtabscale);
 553             ewitab           = _mm_cvttpd_epi32(ewrt);
 554 #ifdef __XOP__
 555             eweps            = _mm_frcz_pd(ewrt);
 556 #else
 557             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 558 #endif
 559             twoeweps         = _mm_add_pd(eweps,eweps);
 560             ewitab           = _mm_slli_epi32(ewitab,2);
 561             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 562             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 563             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 564             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 565             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 566             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 567             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 568             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 569             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
 570             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 571
 572             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velec            = _mm_and_pd(velec,cutoff_mask);
 576             velecsum         = _mm_add_pd(velecsum,velec);
 577
 578             fscal            = felec;
 579
 580             fscal            = _mm_and_pd(fscal,cutoff_mask);
 581
 582             /* Update vectorial force */
 583             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 584             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 585             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 586
 587             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 588             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 589             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 590
 591             }
 592
 593             /**************************
 594              * CALCULATE INTERACTIONS *
 595              **************************/
 596
 597             if (gmx_mm_any_lt(rsq12,rcutoff2))
 598             {
 599
 600             r12              = _mm_mul_pd(rsq12,rinv12);
 601
 602             /* EWALD ELECTROSTATICS */
 603
 604             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 605             ewrt             = _mm_mul_pd(r12,ewtabscale);
 606             ewitab           = _mm_cvttpd_epi32(ewrt);
 607 #ifdef __XOP__
 608             eweps            = _mm_frcz_pd(ewrt);
 609 #else
 610             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 611 #endif
 612             twoeweps         = _mm_add_pd(eweps,eweps);
 613             ewitab           = _mm_slli_epi32(ewitab,2);
 614             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 615             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 616             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 617             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 618             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 619             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 620             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 621             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 622             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
 623             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 624
 625             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             velec            = _mm_and_pd(velec,cutoff_mask);
 629             velecsum         = _mm_add_pd(velecsum,velec);
 630
 631             fscal            = felec;
 632
 633             fscal            = _mm_and_pd(fscal,cutoff_mask);
 634
 635             /* Update vectorial force */
 636             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 637             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 638             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 639
 640             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 641             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 642             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 643
 644             }
 645
 646             /**************************
 647              * CALCULATE INTERACTIONS *
 648              **************************/
 649
 650             if (gmx_mm_any_lt(rsq20,rcutoff2))
 651             {
 652
 653             r20              = _mm_mul_pd(rsq20,rinv20);
 654
 655             /* EWALD ELECTROSTATICS */
 656
 657             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 658             ewrt             = _mm_mul_pd(r20,ewtabscale);
 659             ewitab           = _mm_cvttpd_epi32(ewrt);
 660 #ifdef __XOP__
 661             eweps            = _mm_frcz_pd(ewrt);
 662 #else
 663             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 664 #endif
 665             twoeweps         = _mm_add_pd(eweps,eweps);
 666             ewitab           = _mm_slli_epi32(ewitab,2);
 667             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 668             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 669             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 670             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 671             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 672             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 673             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 674             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 675             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 676             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 677
 678             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 679
 680             /* Update potential sum for this i atom from the interaction with this j atom. */
 681             velec            = _mm_and_pd(velec,cutoff_mask);
 682             velecsum         = _mm_add_pd(velecsum,velec);
 683
 684             fscal            = felec;
 685
 686             fscal            = _mm_and_pd(fscal,cutoff_mask);
 687
 688             /* Update vectorial force */
 689             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 690             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 691             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 692
 693             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 694             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 695             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 696
 697             }
 698
 699             /**************************
 700              * CALCULATE INTERACTIONS *
 701              **************************/
 702
 703             if (gmx_mm_any_lt(rsq21,rcutoff2))
 704             {
 705
 706             r21              = _mm_mul_pd(rsq21,rinv21);
 707
 708             /* EWALD ELECTROSTATICS */
 709
 710             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 711             ewrt             = _mm_mul_pd(r21,ewtabscale);
 712             ewitab           = _mm_cvttpd_epi32(ewrt);
 713 #ifdef __XOP__
 714             eweps            = _mm_frcz_pd(ewrt);
 715 #else
 716             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 717 #endif
 718             twoeweps         = _mm_add_pd(eweps,eweps);
 719             ewitab           = _mm_slli_epi32(ewitab,2);
 720             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 721             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 722             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 723             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 724             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 725             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 726             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 727             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 728             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
 729             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 730
 731             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
 732
 733             /* Update potential sum for this i atom from the interaction with this j atom. */
 734             velec            = _mm_and_pd(velec,cutoff_mask);
 735             velecsum         = _mm_add_pd(velecsum,velec);
 736
 737             fscal            = felec;
 738
 739             fscal            = _mm_and_pd(fscal,cutoff_mask);
 740
 741             /* Update vectorial force */
 742             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 743             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 744             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 745
 746             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 747             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 748             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 749
 750             }
 751
 752             /**************************
 753              * CALCULATE INTERACTIONS *
 754              **************************/
 755
 756             if (gmx_mm_any_lt(rsq22,rcutoff2))
 757             {
 758
 759             r22              = _mm_mul_pd(rsq22,rinv22);
 760
 761             /* EWALD ELECTROSTATICS */
 762
 763             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 764             ewrt             = _mm_mul_pd(r22,ewtabscale);
 765             ewitab           = _mm_cvttpd_epi32(ewrt);
 766 #ifdef __XOP__
 767             eweps            = _mm_frcz_pd(ewrt);
 768 #else
 769             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 770 #endif
 771             twoeweps         = _mm_add_pd(eweps,eweps);
 772             ewitab           = _mm_slli_epi32(ewitab,2);
 773             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 774             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 775             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 776             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 777             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 778             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 779             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 780             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 781             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
 782             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 783
 784             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
 785
 786             /* Update potential sum for this i atom from the interaction with this j atom. */
 787             velec            = _mm_and_pd(velec,cutoff_mask);
 788             velecsum         = _mm_add_pd(velecsum,velec);
 789
 790             fscal            = felec;
 791
 792             fscal            = _mm_and_pd(fscal,cutoff_mask);
 793
 794             /* Update vectorial force */
 795             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 796             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 797             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 798
 799             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 800             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 801             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 802
 803             }
 804
 805             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 806
 807             /* Inner loop uses 470 flops */
 808         }
 809
 810         if(jidx<j_index_end)
 811         {
 812
 813             jnrA             = jjnr[jidx];
 814             j_coord_offsetA  = DIM*jnrA;
 815
 816             /* load j atom coordinates */
 817             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 818                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 819
 820             /* Calculate displacement vector */
 821             dx00             = _mm_sub_pd(ix0,jx0);
 822             dy00             = _mm_sub_pd(iy0,jy0);
 823             dz00             = _mm_sub_pd(iz0,jz0);
 824             dx01             = _mm_sub_pd(ix0,jx1);
 825             dy01             = _mm_sub_pd(iy0,jy1);
 826             dz01             = _mm_sub_pd(iz0,jz1);
 827             dx02             = _mm_sub_pd(ix0,jx2);
 828             dy02             = _mm_sub_pd(iy0,jy2);
 829             dz02             = _mm_sub_pd(iz0,jz2);
 830             dx10             = _mm_sub_pd(ix1,jx0);
 831             dy10             = _mm_sub_pd(iy1,jy0);
 832             dz10             = _mm_sub_pd(iz1,jz0);
 833             dx11             = _mm_sub_pd(ix1,jx1);
 834             dy11             = _mm_sub_pd(iy1,jy1);
 835             dz11             = _mm_sub_pd(iz1,jz1);
 836             dx12             = _mm_sub_pd(ix1,jx2);
 837             dy12             = _mm_sub_pd(iy1,jy2);
 838             dz12             = _mm_sub_pd(iz1,jz2);
 839             dx20             = _mm_sub_pd(ix2,jx0);
 840             dy20             = _mm_sub_pd(iy2,jy0);
 841             dz20             = _mm_sub_pd(iz2,jz0);
 842             dx21             = _mm_sub_pd(ix2,jx1);
 843             dy21             = _mm_sub_pd(iy2,jy1);
 844             dz21             = _mm_sub_pd(iz2,jz1);
 845             dx22             = _mm_sub_pd(ix2,jx2);
 846             dy22             = _mm_sub_pd(iy2,jy2);
 847             dz22             = _mm_sub_pd(iz2,jz2);
 848
 849             /* Calculate squared distance and things based on it */
 850             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 851             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 852             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 853             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 854             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 855             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 856             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 857             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 858             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 859
 860             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 861             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 862             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 863             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 864             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 865             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 866             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 867             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 868             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 869
 870             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 871             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 872             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 873             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 874             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 875             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 876             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 877             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 878             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 879
 880             fjx0             = _mm_setzero_pd();
 881             fjy0             = _mm_setzero_pd();
 882             fjz0             = _mm_setzero_pd();
 883             fjx1             = _mm_setzero_pd();
 884             fjy1             = _mm_setzero_pd();
 885             fjz1             = _mm_setzero_pd();
 886             fjx2             = _mm_setzero_pd();
 887             fjy2             = _mm_setzero_pd();
 888             fjz2             = _mm_setzero_pd();
 889
 890             /**************************
 891              * CALCULATE INTERACTIONS *
 892              **************************/
 893
 894             if (gmx_mm_any_lt(rsq00,rcutoff2))
 895             {
 896
 897             r00              = _mm_mul_pd(rsq00,rinv00);
 898
 899             /* EWALD ELECTROSTATICS */
 900
 901             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 902             ewrt             = _mm_mul_pd(r00,ewtabscale);
 903             ewitab           = _mm_cvttpd_epi32(ewrt);
 904 #ifdef __XOP__
 905             eweps            = _mm_frcz_pd(ewrt);
 906 #else
 907             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 908 #endif
 909             twoeweps         = _mm_add_pd(eweps,eweps);
 910             ewitab           = _mm_slli_epi32(ewitab,2);
 911             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 912             ewtabD           = _mm_setzero_pd();
 913             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 914             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 915             ewtabFn          = _mm_setzero_pd();
 916             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 917             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 918             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 919             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 920             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 921
 922             /* Analytical LJ-PME */
 923             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 924             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 925             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 926             exponent         = gmx_simd_exp_d(ewcljrsq);
 927             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 928             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 929             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 930             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 931             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 932             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 933                                _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));
 934             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 935             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);
 936
 937             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 938
 939             /* Update potential sum for this i atom from the interaction with this j atom. */
 940             velec            = _mm_and_pd(velec,cutoff_mask);
 941             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 942             velecsum         = _mm_add_pd(velecsum,velec);
 943             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 944             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 945             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 946
 947             fscal            = _mm_add_pd(felec,fvdw);
 948
 949             fscal            = _mm_and_pd(fscal,cutoff_mask);
 950
 951             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 952
 953             /* Update vectorial force */
 954             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 955             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 956             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 957
 958             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 959             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 960             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 961
 962             }
 963
 964             /**************************
 965              * CALCULATE INTERACTIONS *
 966              **************************/
 967
 968             if (gmx_mm_any_lt(rsq01,rcutoff2))
 969             {
 970
 971             r01              = _mm_mul_pd(rsq01,rinv01);
 972
 973             /* EWALD ELECTROSTATICS */
 974
 975             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 976             ewrt             = _mm_mul_pd(r01,ewtabscale);
 977             ewitab           = _mm_cvttpd_epi32(ewrt);
 978 #ifdef __XOP__
 979             eweps            = _mm_frcz_pd(ewrt);
 980 #else
 981             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 982 #endif
 983             twoeweps         = _mm_add_pd(eweps,eweps);
 984             ewitab           = _mm_slli_epi32(ewitab,2);
 985             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 986             ewtabD           = _mm_setzero_pd();
 987             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 988             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 989             ewtabFn          = _mm_setzero_pd();
 990             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 991             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 992             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 993             velec            = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
 994             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 995
 996             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
 997
 998             /* Update potential sum for this i atom from the interaction with this j atom. */
 999             velec            = _mm_and_pd(velec,cutoff_mask);
1000             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1001             velecsum         = _mm_add_pd(velecsum,velec);
1002
1003             fscal            = felec;
1004
1005             fscal            = _mm_and_pd(fscal,cutoff_mask);
1006
1007             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1008
1009             /* Update vectorial force */
1010             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1011             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1012             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1013
1014             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1015             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1016             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1017
1018             }
1019
1020             /**************************
1021              * CALCULATE INTERACTIONS *
1022              **************************/
1023
1024             if (gmx_mm_any_lt(rsq02,rcutoff2))
1025             {
1026
1027             r02              = _mm_mul_pd(rsq02,rinv02);
1028
1029             /* EWALD ELECTROSTATICS */
1030
1031             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1032             ewrt             = _mm_mul_pd(r02,ewtabscale);
1033             ewitab           = _mm_cvttpd_epi32(ewrt);
1034 #ifdef __XOP__
1035             eweps            = _mm_frcz_pd(ewrt);
1036 #else
1037             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1038 #endif
1039             twoeweps         = _mm_add_pd(eweps,eweps);
1040             ewitab           = _mm_slli_epi32(ewitab,2);
1041             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1042             ewtabD           = _mm_setzero_pd();
1043             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1044             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1045             ewtabFn          = _mm_setzero_pd();
1046             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1047             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1048             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1049             velec            = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
1050             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1051
1052             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
1053
1054             /* Update potential sum for this i atom from the interaction with this j atom. */
1055             velec            = _mm_and_pd(velec,cutoff_mask);
1056             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1057             velecsum         = _mm_add_pd(velecsum,velec);
1058
1059             fscal            = felec;
1060
1061             fscal            = _mm_and_pd(fscal,cutoff_mask);
1062
1063             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1064
1065             /* Update vectorial force */
1066             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1067             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1068             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1069
1070             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1071             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1072             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1073
1074             }
1075
1076             /**************************
1077              * CALCULATE INTERACTIONS *
1078              **************************/
1079
1080             if (gmx_mm_any_lt(rsq10,rcutoff2))
1081             {
1082
1083             r10              = _mm_mul_pd(rsq10,rinv10);
1084
1085             /* EWALD ELECTROSTATICS */
1086
1087             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1088             ewrt             = _mm_mul_pd(r10,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             ewitab           = _mm_slli_epi32(ewitab,2);
1097             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1098             ewtabD           = _mm_setzero_pd();
1099             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1100             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1101             ewtabFn          = _mm_setzero_pd();
1102             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1103             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1104             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1105             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
1106             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1107
1108             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1109
1110             /* Update potential sum for this i atom from the interaction with this j atom. */
1111             velec            = _mm_and_pd(velec,cutoff_mask);
1112             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1113             velecsum         = _mm_add_pd(velecsum,velec);
1114
1115             fscal            = felec;
1116
1117             fscal            = _mm_and_pd(fscal,cutoff_mask);
1118
1119             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1120
1121             /* Update vectorial force */
1122             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1123             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1124             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1125
1126             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1127             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1128             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1129
1130             }
1131
1132             /**************************
1133              * CALCULATE INTERACTIONS *
1134              **************************/
1135
1136             if (gmx_mm_any_lt(rsq11,rcutoff2))
1137             {
1138
1139             r11              = _mm_mul_pd(rsq11,rinv11);
1140
1141             /* EWALD ELECTROSTATICS */
1142
1143             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1144             ewrt             = _mm_mul_pd(r11,ewtabscale);
1145             ewitab           = _mm_cvttpd_epi32(ewrt);
1146 #ifdef __XOP__
1147             eweps            = _mm_frcz_pd(ewrt);
1148 #else
1149             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1150 #endif
1151             twoeweps         = _mm_add_pd(eweps,eweps);
1152             ewitab           = _mm_slli_epi32(ewitab,2);
1153             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1154             ewtabD           = _mm_setzero_pd();
1155             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1156             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1157             ewtabFn          = _mm_setzero_pd();
1158             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1159             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1160             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1161             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
1162             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1163
1164             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
1165
1166             /* Update potential sum for this i atom from the interaction with this j atom. */
1167             velec            = _mm_and_pd(velec,cutoff_mask);
1168             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1169             velecsum         = _mm_add_pd(velecsum,velec);
1170
1171             fscal            = felec;
1172
1173             fscal            = _mm_and_pd(fscal,cutoff_mask);
1174
1175             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1176
1177             /* Update vectorial force */
1178             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1179             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1180             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1181
1182             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1183             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1184             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1185
1186             }
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             if (gmx_mm_any_lt(rsq12,rcutoff2))
1193             {
1194
1195             r12              = _mm_mul_pd(rsq12,rinv12);
1196
1197             /* EWALD ELECTROSTATICS */
1198
1199             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1200             ewrt             = _mm_mul_pd(r12,ewtabscale);
1201             ewitab           = _mm_cvttpd_epi32(ewrt);
1202 #ifdef __XOP__
1203             eweps            = _mm_frcz_pd(ewrt);
1204 #else
1205             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1206 #endif
1207             twoeweps         = _mm_add_pd(eweps,eweps);
1208             ewitab           = _mm_slli_epi32(ewitab,2);
1209             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1210             ewtabD           = _mm_setzero_pd();
1211             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1212             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1213             ewtabFn          = _mm_setzero_pd();
1214             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1215             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1216             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1217             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
1218             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1219
1220             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
1221
1222             /* Update potential sum for this i atom from the interaction with this j atom. */
1223             velec            = _mm_and_pd(velec,cutoff_mask);
1224             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1225             velecsum         = _mm_add_pd(velecsum,velec);
1226
1227             fscal            = felec;
1228
1229             fscal            = _mm_and_pd(fscal,cutoff_mask);
1230
1231             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1232
1233             /* Update vectorial force */
1234             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1235             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1236             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1237
1238             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1239             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1240             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1241
1242             }
1243
1244             /**************************
1245              * CALCULATE INTERACTIONS *
1246              **************************/
1247
1248             if (gmx_mm_any_lt(rsq20,rcutoff2))
1249             {
1250
1251             r20              = _mm_mul_pd(rsq20,rinv20);
1252
1253             /* EWALD ELECTROSTATICS */
1254
1255             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1256             ewrt             = _mm_mul_pd(r20,ewtabscale);
1257             ewitab           = _mm_cvttpd_epi32(ewrt);
1258 #ifdef __XOP__
1259             eweps            = _mm_frcz_pd(ewrt);
1260 #else
1261             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1262 #endif
1263             twoeweps         = _mm_add_pd(eweps,eweps);
1264             ewitab           = _mm_slli_epi32(ewitab,2);
1265             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1266             ewtabD           = _mm_setzero_pd();
1267             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1268             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1269             ewtabFn          = _mm_setzero_pd();
1270             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1271             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1272             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1273             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
1274             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1275
1276             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1277
1278             /* Update potential sum for this i atom from the interaction with this j atom. */
1279             velec            = _mm_and_pd(velec,cutoff_mask);
1280             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1281             velecsum         = _mm_add_pd(velecsum,velec);
1282
1283             fscal            = felec;
1284
1285             fscal            = _mm_and_pd(fscal,cutoff_mask);
1286
1287             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1288
1289             /* Update vectorial force */
1290             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1291             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1292             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1293
1294             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1295             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1296             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1297
1298             }
1299
1300             /**************************
1301              * CALCULATE INTERACTIONS *
1302              **************************/
1303
1304             if (gmx_mm_any_lt(rsq21,rcutoff2))
1305             {
1306
1307             r21              = _mm_mul_pd(rsq21,rinv21);
1308
1309             /* EWALD ELECTROSTATICS */
1310
1311             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1312             ewrt             = _mm_mul_pd(r21,ewtabscale);
1313             ewitab           = _mm_cvttpd_epi32(ewrt);
1314 #ifdef __XOP__
1315             eweps            = _mm_frcz_pd(ewrt);
1316 #else
1317             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1318 #endif
1319             twoeweps         = _mm_add_pd(eweps,eweps);
1320             ewitab           = _mm_slli_epi32(ewitab,2);
1321             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1322             ewtabD           = _mm_setzero_pd();
1323             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1324             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1325             ewtabFn          = _mm_setzero_pd();
1326             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1327             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1328             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1329             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
1330             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1331
1332             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
1333
1334             /* Update potential sum for this i atom from the interaction with this j atom. */
1335             velec            = _mm_and_pd(velec,cutoff_mask);
1336             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1337             velecsum         = _mm_add_pd(velecsum,velec);
1338
1339             fscal            = felec;
1340
1341             fscal            = _mm_and_pd(fscal,cutoff_mask);
1342
1343             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1344
1345             /* Update vectorial force */
1346             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1347             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1348             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1349
1350             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1351             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1352             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1353
1354             }
1355
1356             /**************************
1357              * CALCULATE INTERACTIONS *
1358              **************************/
1359
1360             if (gmx_mm_any_lt(rsq22,rcutoff2))
1361             {
1362
1363             r22              = _mm_mul_pd(rsq22,rinv22);
1364
1365             /* EWALD ELECTROSTATICS */
1366
1367             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1368             ewrt             = _mm_mul_pd(r22,ewtabscale);
1369             ewitab           = _mm_cvttpd_epi32(ewrt);
1370 #ifdef __XOP__
1371             eweps            = _mm_frcz_pd(ewrt);
1372 #else
1373             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1374 #endif
1375             twoeweps         = _mm_add_pd(eweps,eweps);
1376             ewitab           = _mm_slli_epi32(ewitab,2);
1377             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1378             ewtabD           = _mm_setzero_pd();
1379             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1380             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1381             ewtabFn          = _mm_setzero_pd();
1382             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1383             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1384             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1385             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
1386             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1387
1388             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
1389
1390             /* Update potential sum for this i atom from the interaction with this j atom. */
1391             velec            = _mm_and_pd(velec,cutoff_mask);
1392             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1393             velecsum         = _mm_add_pd(velecsum,velec);
1394
1395             fscal            = felec;
1396
1397             fscal            = _mm_and_pd(fscal,cutoff_mask);
1398
1399             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1400
1401             /* Update vectorial force */
1402             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1403             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1404             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1405
1406             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1407             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1408             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1409
1410             }
1411
1412             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1413
1414             /* Inner loop uses 470 flops */
1415         }
1416
1417         /* End of innermost loop */
1418
1419         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1420                                               f+i_coord_offset,fshift+i_shift_offset);
1421
1422         ggid                        = gid[iidx];
1423         /* Update potential energies */
1424         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1425         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1426
1427         /* Increment number of inner iterations */
1428         inneriter                  += j_index_end - j_index_start;
1429
1430         /* Outer loop uses 20 flops */
1431     }
1432
1433     /* Increment number of outer iterations */
1434     outeriter        += nri;
1435
1436     /* Update outer/inner flops */
1437
1438     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*470);
1439 }
1440 /*
1441  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_F_avx_128_fma_double
1442  * Electrostatics interaction: Ewald
1443  * VdW interaction:            LJEwald
1444  * Geometry:                   Water3-Water3
1445  * Calculate force/pot:        Force
1446  */
1447 void
1448 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3W3_F_avx_128_fma_double
1449                     (t_nblist                    * gmx_restrict       nlist,
1450                      rvec                        * gmx_restrict          xx,
1451                      rvec                        * gmx_restrict          ff,
1452                      t_forcerec                  * gmx_restrict          fr,
1453                      t_mdatoms                   * gmx_restrict     mdatoms,
1454                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1455                      t_nrnb                      * gmx_restrict        nrnb)
1456 {
1457     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1458      * just 0 for non-waters.
1459      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1460      * jnr indices corresponding to data put in the four positions in the SIMD register.
1461      */
1462     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1463     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1464     int              jnrA,jnrB;
1465     int              j_coord_offsetA,j_coord_offsetB;
1466     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1467     real             rcutoff_scalar;
1468     real             *shiftvec,*fshift,*x,*f;
1469     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1470     int              vdwioffset0;
1471     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1472     int              vdwioffset1;
1473     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1474     int              vdwioffset2;
1475     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1476     int              vdwjidx0A,vdwjidx0B;
1477     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1478     int              vdwjidx1A,vdwjidx1B;
1479     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1480     int              vdwjidx2A,vdwjidx2B;
1481     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1482     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1483     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1484     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1485     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1486     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1487     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1488     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1489     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1490     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1491     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1492     real             *charge;
1493     int              nvdwtype;
1494     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1495     int              *vdwtype;
1496     real             *vdwparam;
1497     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1498     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1499     __m128d           c6grid_00;
1500     __m128d           c6grid_01;
1501     __m128d           c6grid_02;
1502     __m128d           c6grid_10;
1503     __m128d           c6grid_11;
1504     __m128d           c6grid_12;
1505     __m128d           c6grid_20;
1506     __m128d           c6grid_21;
1507     __m128d           c6grid_22;
1508     real             *vdwgridparam;
1509     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1510     __m128d           one_half  = _mm_set1_pd(0.5);
1511     __m128d           minus_one = _mm_set1_pd(-1.0);
1512     __m128i          ewitab;
1513     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1514     real             *ewtab;
1515     __m128d          dummy_mask,cutoff_mask;
1516     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1517     __m128d          one     = _mm_set1_pd(1.0);
1518     __m128d          two     = _mm_set1_pd(2.0);
1519     x                = xx[0];
1520     f                = ff[0];
1521
1522     nri              = nlist->nri;
1523     iinr             = nlist->iinr;
1524     jindex           = nlist->jindex;
1525     jjnr             = nlist->jjnr;
1526     shiftidx         = nlist->shift;
1527     gid              = nlist->gid;
1528     shiftvec         = fr->shift_vec[0];
1529     fshift           = fr->fshift[0];
1530     facel            = _mm_set1_pd(fr->epsfac);
1531     charge           = mdatoms->chargeA;
1532     nvdwtype         = fr->ntype;
1533     vdwparam         = fr->nbfp;
1534     vdwtype          = mdatoms->typeA;
1535     vdwgridparam     = fr->ljpme_c6grid;
1536     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
1537     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
1538     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
1539
1540     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1541     ewtab            = fr->ic->tabq_coul_F;
1542     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1543     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1544
1545     /* Setup water-specific parameters */
1546     inr              = nlist->iinr[0];
1547     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1548     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1549     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1550     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1551
1552     jq0              = _mm_set1_pd(charge[inr+0]);
1553     jq1              = _mm_set1_pd(charge[inr+1]);
1554     jq2              = _mm_set1_pd(charge[inr+2]);
1555     vdwjidx0A        = 2*vdwtype[inr+0];
1556     qq00             = _mm_mul_pd(iq0,jq0);
1557     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1558     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1559     c6grid_00        = _mm_set1_pd(vdwgridparam[vdwioffset0+vdwjidx0A]);
1560     qq01             = _mm_mul_pd(iq0,jq1);
1561     qq02             = _mm_mul_pd(iq0,jq2);
1562     qq10             = _mm_mul_pd(iq1,jq0);
1563     qq11             = _mm_mul_pd(iq1,jq1);
1564     qq12             = _mm_mul_pd(iq1,jq2);
1565     qq20             = _mm_mul_pd(iq2,jq0);
1566     qq21             = _mm_mul_pd(iq2,jq1);
1567     qq22             = _mm_mul_pd(iq2,jq2);
1568
1569     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1570     rcutoff_scalar   = fr->rcoulomb;
1571     rcutoff          = _mm_set1_pd(rcutoff_scalar);
1572     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
1573
1574     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
1575     rvdw             = _mm_set1_pd(fr->rvdw);
1576
1577     /* Avoid stupid compiler warnings */
1578     jnrA = jnrB = 0;
1579     j_coord_offsetA = 0;
1580     j_coord_offsetB = 0;
1581
1582     outeriter        = 0;
1583     inneriter        = 0;
1584
1585     /* Start outer loop over neighborlists */
1586     for(iidx=0; iidx<nri; iidx++)
1587     {
1588         /* Load shift vector for this list */
1589         i_shift_offset   = DIM*shiftidx[iidx];
1590
1591         /* Load limits for loop over neighbors */
1592         j_index_start    = jindex[iidx];
1593         j_index_end      = jindex[iidx+1];
1594
1595         /* Get outer coordinate index */
1596         inr              = iinr[iidx];
1597         i_coord_offset   = DIM*inr;
1598
1599         /* Load i particle coords and add shift vector */
1600         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1601                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1602
1603         fix0             = _mm_setzero_pd();
1604         fiy0             = _mm_setzero_pd();
1605         fiz0             = _mm_setzero_pd();
1606         fix1             = _mm_setzero_pd();
1607         fiy1             = _mm_setzero_pd();
1608         fiz1             = _mm_setzero_pd();
1609         fix2             = _mm_setzero_pd();
1610         fiy2             = _mm_setzero_pd();
1611         fiz2             = _mm_setzero_pd();
1612
1613         /* Start inner kernel loop */
1614         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1615         {
1616
1617             /* Get j neighbor index, and coordinate index */
1618             jnrA             = jjnr[jidx];
1619             jnrB             = jjnr[jidx+1];
1620             j_coord_offsetA  = DIM*jnrA;
1621             j_coord_offsetB  = DIM*jnrB;
1622
1623             /* load j atom coordinates */
1624             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1625                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1626
1627             /* Calculate displacement vector */
1628             dx00             = _mm_sub_pd(ix0,jx0);
1629             dy00             = _mm_sub_pd(iy0,jy0);
1630             dz00             = _mm_sub_pd(iz0,jz0);
1631             dx01             = _mm_sub_pd(ix0,jx1);
1632             dy01             = _mm_sub_pd(iy0,jy1);
1633             dz01             = _mm_sub_pd(iz0,jz1);
1634             dx02             = _mm_sub_pd(ix0,jx2);
1635             dy02             = _mm_sub_pd(iy0,jy2);
1636             dz02             = _mm_sub_pd(iz0,jz2);
1637             dx10             = _mm_sub_pd(ix1,jx0);
1638             dy10             = _mm_sub_pd(iy1,jy0);
1639             dz10             = _mm_sub_pd(iz1,jz0);
1640             dx11             = _mm_sub_pd(ix1,jx1);
1641             dy11             = _mm_sub_pd(iy1,jy1);
1642             dz11             = _mm_sub_pd(iz1,jz1);
1643             dx12             = _mm_sub_pd(ix1,jx2);
1644             dy12             = _mm_sub_pd(iy1,jy2);
1645             dz12             = _mm_sub_pd(iz1,jz2);
1646             dx20             = _mm_sub_pd(ix2,jx0);
1647             dy20             = _mm_sub_pd(iy2,jy0);
1648             dz20             = _mm_sub_pd(iz2,jz0);
1649             dx21             = _mm_sub_pd(ix2,jx1);
1650             dy21             = _mm_sub_pd(iy2,jy1);
1651             dz21             = _mm_sub_pd(iz2,jz1);
1652             dx22             = _mm_sub_pd(ix2,jx2);
1653             dy22             = _mm_sub_pd(iy2,jy2);
1654             dz22             = _mm_sub_pd(iz2,jz2);
1655
1656             /* Calculate squared distance and things based on it */
1657             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1658             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1659             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1660             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1661             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1662             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1663             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1664             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1665             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1666
1667             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1668             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1669             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1670             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1671             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1672             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1673             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1674             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1675             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1676
1677             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1678             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1679             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1680             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1681             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1682             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1683             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1684             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1685             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1686
1687             fjx0             = _mm_setzero_pd();
1688             fjy0             = _mm_setzero_pd();
1689             fjz0             = _mm_setzero_pd();
1690             fjx1             = _mm_setzero_pd();
1691             fjy1             = _mm_setzero_pd();
1692             fjz1             = _mm_setzero_pd();
1693             fjx2             = _mm_setzero_pd();
1694             fjy2             = _mm_setzero_pd();
1695             fjz2             = _mm_setzero_pd();
1696
1697             /**************************
1698              * CALCULATE INTERACTIONS *
1699              **************************/
1700
1701             if (gmx_mm_any_lt(rsq00,rcutoff2))
1702             {
1703
1704             r00              = _mm_mul_pd(rsq00,rinv00);
1705
1706             /* EWALD ELECTROSTATICS */
1707
1708             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1709             ewrt             = _mm_mul_pd(r00,ewtabscale);
1710             ewitab           = _mm_cvttpd_epi32(ewrt);
1711 #ifdef __XOP__
1712             eweps            = _mm_frcz_pd(ewrt);
1713 #else
1714             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1715 #endif
1716             twoeweps         = _mm_add_pd(eweps,eweps);
1717             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1718                                          &ewtabF,&ewtabFn);
1719             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1720             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1721
1722             /* Analytical LJ-PME */
1723             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1724             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1725             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1726             exponent         = gmx_simd_exp_d(ewcljrsq);
1727             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1728             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1729             /* f6A = 6 * C6grid * (1 - poly) */
1730             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1731             /* f6B = C6grid * exponent * beta^6 */
1732             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1733             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1734             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1735
1736             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1737
1738             fscal            = _mm_add_pd(felec,fvdw);
1739
1740             fscal            = _mm_and_pd(fscal,cutoff_mask);
1741
1742             /* Update vectorial force */
1743             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1744             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1745             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1746
1747             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1748             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1749             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1750
1751             }
1752
1753             /**************************
1754              * CALCULATE INTERACTIONS *
1755              **************************/
1756
1757             if (gmx_mm_any_lt(rsq01,rcutoff2))
1758             {
1759
1760             r01              = _mm_mul_pd(rsq01,rinv01);
1761
1762             /* EWALD ELECTROSTATICS */
1763
1764             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1765             ewrt             = _mm_mul_pd(r01,ewtabscale);
1766             ewitab           = _mm_cvttpd_epi32(ewrt);
1767 #ifdef __XOP__
1768             eweps            = _mm_frcz_pd(ewrt);
1769 #else
1770             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1771 #endif
1772             twoeweps         = _mm_add_pd(eweps,eweps);
1773             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1774                                          &ewtabF,&ewtabFn);
1775             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1776             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1777
1778             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
1779
1780             fscal            = felec;
1781
1782             fscal            = _mm_and_pd(fscal,cutoff_mask);
1783
1784             /* Update vectorial force */
1785             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1786             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1787             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1788
1789             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1790             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1791             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1792
1793             }
1794
1795             /**************************
1796              * CALCULATE INTERACTIONS *
1797              **************************/
1798
1799             if (gmx_mm_any_lt(rsq02,rcutoff2))
1800             {
1801
1802             r02              = _mm_mul_pd(rsq02,rinv02);
1803
1804             /* EWALD ELECTROSTATICS */
1805
1806             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1807             ewrt             = _mm_mul_pd(r02,ewtabscale);
1808             ewitab           = _mm_cvttpd_epi32(ewrt);
1809 #ifdef __XOP__
1810             eweps            = _mm_frcz_pd(ewrt);
1811 #else
1812             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1813 #endif
1814             twoeweps         = _mm_add_pd(eweps,eweps);
1815             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1816                                          &ewtabF,&ewtabFn);
1817             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1818             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1819
1820             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
1821
1822             fscal            = felec;
1823
1824             fscal            = _mm_and_pd(fscal,cutoff_mask);
1825
1826             /* Update vectorial force */
1827             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1828             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1829             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1830
1831             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1832             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1833             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1834
1835             }
1836
1837             /**************************
1838              * CALCULATE INTERACTIONS *
1839              **************************/
1840
1841             if (gmx_mm_any_lt(rsq10,rcutoff2))
1842             {
1843
1844             r10              = _mm_mul_pd(rsq10,rinv10);
1845
1846             /* EWALD ELECTROSTATICS */
1847
1848             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1849             ewrt             = _mm_mul_pd(r10,ewtabscale);
1850             ewitab           = _mm_cvttpd_epi32(ewrt);
1851 #ifdef __XOP__
1852             eweps            = _mm_frcz_pd(ewrt);
1853 #else
1854             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1855 #endif
1856             twoeweps         = _mm_add_pd(eweps,eweps);
1857             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1858                                          &ewtabF,&ewtabFn);
1859             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1860             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1861
1862             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1863
1864             fscal            = felec;
1865
1866             fscal            = _mm_and_pd(fscal,cutoff_mask);
1867
1868             /* Update vectorial force */
1869             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1870             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1871             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1872
1873             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1874             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1875             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1876
1877             }
1878
1879             /**************************
1880              * CALCULATE INTERACTIONS *
1881              **************************/
1882
1883             if (gmx_mm_any_lt(rsq11,rcutoff2))
1884             {
1885
1886             r11              = _mm_mul_pd(rsq11,rinv11);
1887
1888             /* EWALD ELECTROSTATICS */
1889
1890             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1891             ewrt             = _mm_mul_pd(r11,ewtabscale);
1892             ewitab           = _mm_cvttpd_epi32(ewrt);
1893 #ifdef __XOP__
1894             eweps            = _mm_frcz_pd(ewrt);
1895 #else
1896             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1897 #endif
1898             twoeweps         = _mm_add_pd(eweps,eweps);
1899             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1900                                          &ewtabF,&ewtabFn);
1901             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1902             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1903
1904             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
1905
1906             fscal            = felec;
1907
1908             fscal            = _mm_and_pd(fscal,cutoff_mask);
1909
1910             /* Update vectorial force */
1911             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1912             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1913             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1914
1915             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1916             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1917             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1918
1919             }
1920
1921             /**************************
1922              * CALCULATE INTERACTIONS *
1923              **************************/
1924
1925             if (gmx_mm_any_lt(rsq12,rcutoff2))
1926             {
1927
1928             r12              = _mm_mul_pd(rsq12,rinv12);
1929
1930             /* EWALD ELECTROSTATICS */
1931
1932             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1933             ewrt             = _mm_mul_pd(r12,ewtabscale);
1934             ewitab           = _mm_cvttpd_epi32(ewrt);
1935 #ifdef __XOP__
1936             eweps            = _mm_frcz_pd(ewrt);
1937 #else
1938             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1939 #endif
1940             twoeweps         = _mm_add_pd(eweps,eweps);
1941             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1942                                          &ewtabF,&ewtabFn);
1943             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1944             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1945
1946             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
1947
1948             fscal            = felec;
1949
1950             fscal            = _mm_and_pd(fscal,cutoff_mask);
1951
1952             /* Update vectorial force */
1953             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1954             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1955             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1956
1957             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1958             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1959             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1960
1961             }
1962
1963             /**************************
1964              * CALCULATE INTERACTIONS *
1965              **************************/
1966
1967             if (gmx_mm_any_lt(rsq20,rcutoff2))
1968             {
1969
1970             r20              = _mm_mul_pd(rsq20,rinv20);
1971
1972             /* EWALD ELECTROSTATICS */
1973
1974             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1975             ewrt             = _mm_mul_pd(r20,ewtabscale);
1976             ewitab           = _mm_cvttpd_epi32(ewrt);
1977 #ifdef __XOP__
1978             eweps            = _mm_frcz_pd(ewrt);
1979 #else
1980             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1981 #endif
1982             twoeweps         = _mm_add_pd(eweps,eweps);
1983             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1984                                          &ewtabF,&ewtabFn);
1985             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1986             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1987
1988             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1989
1990             fscal            = felec;
1991
1992             fscal            = _mm_and_pd(fscal,cutoff_mask);
1993
1994             /* Update vectorial force */
1995             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1996             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1997             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1998
1999             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2000             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2001             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
2002
2003             }
2004
2005             /**************************
2006              * CALCULATE INTERACTIONS *
2007              **************************/
2008
2009             if (gmx_mm_any_lt(rsq21,rcutoff2))
2010             {
2011
2012             r21              = _mm_mul_pd(rsq21,rinv21);
2013
2014             /* EWALD ELECTROSTATICS */
2015
2016             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2017             ewrt             = _mm_mul_pd(r21,ewtabscale);
2018             ewitab           = _mm_cvttpd_epi32(ewrt);
2019 #ifdef __XOP__
2020             eweps            = _mm_frcz_pd(ewrt);
2021 #else
2022             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2023 #endif
2024             twoeweps         = _mm_add_pd(eweps,eweps);
2025             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
2026                                          &ewtabF,&ewtabFn);
2027             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2028             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2029
2030             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
2031
2032             fscal            = felec;
2033
2034             fscal            = _mm_and_pd(fscal,cutoff_mask);
2035
2036             /* Update vectorial force */
2037             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2038             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2039             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2040
2041             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2042             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2043             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2044
2045             }
2046
2047             /**************************
2048              * CALCULATE INTERACTIONS *
2049              **************************/
2050
2051             if (gmx_mm_any_lt(rsq22,rcutoff2))
2052             {
2053
2054             r22              = _mm_mul_pd(rsq22,rinv22);
2055
2056             /* EWALD ELECTROSTATICS */
2057
2058             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2059             ewrt             = _mm_mul_pd(r22,ewtabscale);
2060             ewitab           = _mm_cvttpd_epi32(ewrt);
2061 #ifdef __XOP__
2062             eweps            = _mm_frcz_pd(ewrt);
2063 #else
2064             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2065 #endif
2066             twoeweps         = _mm_add_pd(eweps,eweps);
2067             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
2068                                          &ewtabF,&ewtabFn);
2069             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2070             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2071
2072             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
2073
2074             fscal            = felec;
2075
2076             fscal            = _mm_and_pd(fscal,cutoff_mask);
2077
2078             /* Update vectorial force */
2079             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2080             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2081             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2082
2083             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2084             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2085             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2086
2087             }
2088
2089             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2090
2091             /* Inner loop uses 399 flops */
2092         }
2093
2094         if(jidx<j_index_end)
2095         {
2096
2097             jnrA             = jjnr[jidx];
2098             j_coord_offsetA  = DIM*jnrA;
2099
2100             /* load j atom coordinates */
2101             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
2102                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2103
2104             /* Calculate displacement vector */
2105             dx00             = _mm_sub_pd(ix0,jx0);
2106             dy00             = _mm_sub_pd(iy0,jy0);
2107             dz00             = _mm_sub_pd(iz0,jz0);
2108             dx01             = _mm_sub_pd(ix0,jx1);
2109             dy01             = _mm_sub_pd(iy0,jy1);
2110             dz01             = _mm_sub_pd(iz0,jz1);
2111             dx02             = _mm_sub_pd(ix0,jx2);
2112             dy02             = _mm_sub_pd(iy0,jy2);
2113             dz02             = _mm_sub_pd(iz0,jz2);
2114             dx10             = _mm_sub_pd(ix1,jx0);
2115             dy10             = _mm_sub_pd(iy1,jy0);
2116             dz10             = _mm_sub_pd(iz1,jz0);
2117             dx11             = _mm_sub_pd(ix1,jx1);
2118             dy11             = _mm_sub_pd(iy1,jy1);
2119             dz11             = _mm_sub_pd(iz1,jz1);
2120             dx12             = _mm_sub_pd(ix1,jx2);
2121             dy12             = _mm_sub_pd(iy1,jy2);
2122             dz12             = _mm_sub_pd(iz1,jz2);
2123             dx20             = _mm_sub_pd(ix2,jx0);
2124             dy20             = _mm_sub_pd(iy2,jy0);
2125             dz20             = _mm_sub_pd(iz2,jz0);
2126             dx21             = _mm_sub_pd(ix2,jx1);
2127             dy21             = _mm_sub_pd(iy2,jy1);
2128             dz21             = _mm_sub_pd(iz2,jz1);
2129             dx22             = _mm_sub_pd(ix2,jx2);
2130             dy22             = _mm_sub_pd(iy2,jy2);
2131             dz22             = _mm_sub_pd(iz2,jz2);
2132
2133             /* Calculate squared distance and things based on it */
2134             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2135             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
2136             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
2137             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
2138             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2139             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2140             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
2141             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2142             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2143
2144             rinv00           = gmx_mm_invsqrt_pd(rsq00);
2145             rinv01           = gmx_mm_invsqrt_pd(rsq01);
2146             rinv02           = gmx_mm_invsqrt_pd(rsq02);
2147             rinv10           = gmx_mm_invsqrt_pd(rsq10);
2148             rinv11           = gmx_mm_invsqrt_pd(rsq11);
2149             rinv12           = gmx_mm_invsqrt_pd(rsq12);
2150             rinv20           = gmx_mm_invsqrt_pd(rsq20);
2151             rinv21           = gmx_mm_invsqrt_pd(rsq21);
2152             rinv22           = gmx_mm_invsqrt_pd(rsq22);
2153
2154             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
2155             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
2156             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
2157             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
2158             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
2159             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
2160             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
2161             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
2162             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
2163
2164             fjx0             = _mm_setzero_pd();
2165             fjy0             = _mm_setzero_pd();
2166             fjz0             = _mm_setzero_pd();
2167             fjx1             = _mm_setzero_pd();
2168             fjy1             = _mm_setzero_pd();
2169             fjz1             = _mm_setzero_pd();
2170             fjx2             = _mm_setzero_pd();
2171             fjy2             = _mm_setzero_pd();
2172             fjz2             = _mm_setzero_pd();
2173
2174             /**************************
2175              * CALCULATE INTERACTIONS *
2176              **************************/
2177
2178             if (gmx_mm_any_lt(rsq00,rcutoff2))
2179             {
2180
2181             r00              = _mm_mul_pd(rsq00,rinv00);
2182
2183             /* EWALD ELECTROSTATICS */
2184
2185             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2186             ewrt             = _mm_mul_pd(r00,ewtabscale);
2187             ewitab           = _mm_cvttpd_epi32(ewrt);
2188 #ifdef __XOP__
2189             eweps            = _mm_frcz_pd(ewrt);
2190 #else
2191             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2192 #endif
2193             twoeweps         = _mm_add_pd(eweps,eweps);
2194             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2195             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2196             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
2197
2198             /* Analytical LJ-PME */
2199             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2200             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
2201             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
2202             exponent         = gmx_simd_exp_d(ewcljrsq);
2203             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2204             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
2205             /* f6A = 6 * C6grid * (1 - poly) */
2206             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
2207             /* f6B = C6grid * exponent * beta^6 */
2208             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
2209             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2210             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
2211
2212             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
2213
2214             fscal            = _mm_add_pd(felec,fvdw);
2215
2216             fscal            = _mm_and_pd(fscal,cutoff_mask);
2217
2218             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2219
2220             /* Update vectorial force */
2221             fix0             = _mm_macc_pd(dx00,fscal,fix0);
2222             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
2223             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
2224
2225             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
2226             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
2227             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
2228
2229             }
2230
2231             /**************************
2232              * CALCULATE INTERACTIONS *
2233              **************************/
2234
2235             if (gmx_mm_any_lt(rsq01,rcutoff2))
2236             {
2237
2238             r01              = _mm_mul_pd(rsq01,rinv01);
2239
2240             /* EWALD ELECTROSTATICS */
2241
2242             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2243             ewrt             = _mm_mul_pd(r01,ewtabscale);
2244             ewitab           = _mm_cvttpd_epi32(ewrt);
2245 #ifdef __XOP__
2246             eweps            = _mm_frcz_pd(ewrt);
2247 #else
2248             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2249 #endif
2250             twoeweps         = _mm_add_pd(eweps,eweps);
2251             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2252             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2253             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
2254
2255             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
2256
2257             fscal            = felec;
2258
2259             fscal            = _mm_and_pd(fscal,cutoff_mask);
2260
2261             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2262
2263             /* Update vectorial force */
2264             fix0             = _mm_macc_pd(dx01,fscal,fix0);
2265             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
2266             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
2267
2268             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
2269             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
2270             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
2271
2272             }
2273
2274             /**************************
2275              * CALCULATE INTERACTIONS *
2276              **************************/
2277
2278             if (gmx_mm_any_lt(rsq02,rcutoff2))
2279             {
2280
2281             r02              = _mm_mul_pd(rsq02,rinv02);
2282
2283             /* EWALD ELECTROSTATICS */
2284
2285             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2286             ewrt             = _mm_mul_pd(r02,ewtabscale);
2287             ewitab           = _mm_cvttpd_epi32(ewrt);
2288 #ifdef __XOP__
2289             eweps            = _mm_frcz_pd(ewrt);
2290 #else
2291             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2292 #endif
2293             twoeweps         = _mm_add_pd(eweps,eweps);
2294             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2295             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2296             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
2297
2298             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
2299
2300             fscal            = felec;
2301
2302             fscal            = _mm_and_pd(fscal,cutoff_mask);
2303
2304             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2305
2306             /* Update vectorial force */
2307             fix0             = _mm_macc_pd(dx02,fscal,fix0);
2308             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
2309             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
2310
2311             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
2312             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
2313             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
2314
2315             }
2316
2317             /**************************
2318              * CALCULATE INTERACTIONS *
2319              **************************/
2320
2321             if (gmx_mm_any_lt(rsq10,rcutoff2))
2322             {
2323
2324             r10              = _mm_mul_pd(rsq10,rinv10);
2325
2326             /* EWALD ELECTROSTATICS */
2327
2328             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2329             ewrt             = _mm_mul_pd(r10,ewtabscale);
2330             ewitab           = _mm_cvttpd_epi32(ewrt);
2331 #ifdef __XOP__
2332             eweps            = _mm_frcz_pd(ewrt);
2333 #else
2334             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2335 #endif
2336             twoeweps         = _mm_add_pd(eweps,eweps);
2337             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2338             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2339             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2340
2341             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
2342
2343             fscal            = felec;
2344
2345             fscal            = _mm_and_pd(fscal,cutoff_mask);
2346
2347             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2348
2349             /* Update vectorial force */
2350             fix1             = _mm_macc_pd(dx10,fscal,fix1);
2351             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
2352             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
2353
2354             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
2355             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
2356             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
2357
2358             }
2359
2360             /**************************
2361              * CALCULATE INTERACTIONS *
2362              **************************/
2363
2364             if (gmx_mm_any_lt(rsq11,rcutoff2))
2365             {
2366
2367             r11              = _mm_mul_pd(rsq11,rinv11);
2368
2369             /* EWALD ELECTROSTATICS */
2370
2371             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2372             ewrt             = _mm_mul_pd(r11,ewtabscale);
2373             ewitab           = _mm_cvttpd_epi32(ewrt);
2374 #ifdef __XOP__
2375             eweps            = _mm_frcz_pd(ewrt);
2376 #else
2377             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2378 #endif
2379             twoeweps         = _mm_add_pd(eweps,eweps);
2380             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2381             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2382             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2383
2384             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
2385
2386             fscal            = felec;
2387
2388             fscal            = _mm_and_pd(fscal,cutoff_mask);
2389
2390             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2391
2392             /* Update vectorial force */
2393             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2394             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2395             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2396
2397             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2398             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2399             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2400
2401             }
2402
2403             /**************************
2404              * CALCULATE INTERACTIONS *
2405              **************************/
2406
2407             if (gmx_mm_any_lt(rsq12,rcutoff2))
2408             {
2409
2410             r12              = _mm_mul_pd(rsq12,rinv12);
2411
2412             /* EWALD ELECTROSTATICS */
2413
2414             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2415             ewrt             = _mm_mul_pd(r12,ewtabscale);
2416             ewitab           = _mm_cvttpd_epi32(ewrt);
2417 #ifdef __XOP__
2418             eweps            = _mm_frcz_pd(ewrt);
2419 #else
2420             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2421 #endif
2422             twoeweps         = _mm_add_pd(eweps,eweps);
2423             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2424             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2425             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2426
2427             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
2428
2429             fscal            = felec;
2430
2431             fscal            = _mm_and_pd(fscal,cutoff_mask);
2432
2433             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2434
2435             /* Update vectorial force */
2436             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2437             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2438             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2439
2440             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2441             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2442             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2443
2444             }
2445
2446             /**************************
2447              * CALCULATE INTERACTIONS *
2448              **************************/
2449
2450             if (gmx_mm_any_lt(rsq20,rcutoff2))
2451             {
2452
2453             r20              = _mm_mul_pd(rsq20,rinv20);
2454
2455             /* EWALD ELECTROSTATICS */
2456
2457             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2458             ewrt             = _mm_mul_pd(r20,ewtabscale);
2459             ewitab           = _mm_cvttpd_epi32(ewrt);
2460 #ifdef __XOP__
2461             eweps            = _mm_frcz_pd(ewrt);
2462 #else
2463             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2464 #endif
2465             twoeweps         = _mm_add_pd(eweps,eweps);
2466             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2467             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2468             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2469
2470             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
2471
2472             fscal            = felec;
2473
2474             fscal            = _mm_and_pd(fscal,cutoff_mask);
2475
2476             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2477
2478             /* Update vectorial force */
2479             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2480             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2481             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2482
2483             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2484             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2485             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
2486
2487             }
2488
2489             /**************************
2490              * CALCULATE INTERACTIONS *
2491              **************************/
2492
2493             if (gmx_mm_any_lt(rsq21,rcutoff2))
2494             {
2495
2496             r21              = _mm_mul_pd(rsq21,rinv21);
2497
2498             /* EWALD ELECTROSTATICS */
2499
2500             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2501             ewrt             = _mm_mul_pd(r21,ewtabscale);
2502             ewitab           = _mm_cvttpd_epi32(ewrt);
2503 #ifdef __XOP__
2504             eweps            = _mm_frcz_pd(ewrt);
2505 #else
2506             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2507 #endif
2508             twoeweps         = _mm_add_pd(eweps,eweps);
2509             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2510             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2511             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2512
2513             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
2514
2515             fscal            = felec;
2516
2517             fscal            = _mm_and_pd(fscal,cutoff_mask);
2518
2519             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2520
2521             /* Update vectorial force */
2522             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2523             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2524             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2525
2526             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2527             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2528             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2529
2530             }
2531
2532             /**************************
2533              * CALCULATE INTERACTIONS *
2534              **************************/
2535
2536             if (gmx_mm_any_lt(rsq22,rcutoff2))
2537             {
2538
2539             r22              = _mm_mul_pd(rsq22,rinv22);
2540
2541             /* EWALD ELECTROSTATICS */
2542
2543             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2544             ewrt             = _mm_mul_pd(r22,ewtabscale);
2545             ewitab           = _mm_cvttpd_epi32(ewrt);
2546 #ifdef __XOP__
2547             eweps            = _mm_frcz_pd(ewrt);
2548 #else
2549             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2550 #endif
2551             twoeweps         = _mm_add_pd(eweps,eweps);
2552             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2553             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2554             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2555
2556             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
2557
2558             fscal            = felec;
2559
2560             fscal            = _mm_and_pd(fscal,cutoff_mask);
2561
2562             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2563
2564             /* Update vectorial force */
2565             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2566             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2567             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2568
2569             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2570             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2571             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2572
2573             }
2574
2575             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2576
2577             /* Inner loop uses 399 flops */
2578         }
2579
2580         /* End of innermost loop */
2581
2582         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2583                                               f+i_coord_offset,fshift+i_shift_offset);
2584
2585         /* Increment number of inner iterations */
2586         inneriter                  += j_index_end - j_index_start;
2587
2588         /* Outer loop uses 18 flops */
2589     }
2590
2591     /* Increment number of outer iterations */
2592     outeriter        += nri;
2593
2594     /* Update outer/inner flops */
2595
2596     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*399);
2597 }