src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Water4
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     int              vdwjidx1A,vdwjidx1B;
  90     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  91     int              vdwjidx2A,vdwjidx2B;
  92     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  93     int              vdwjidx3A,vdwjidx3B;
  94     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  95     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  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          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  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          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 102     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 103     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 104     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 105     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 106     real             *charge;
 107     int              nvdwtype;
 108     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 109     int              *vdwtype;
 110     real             *vdwparam;
 111     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 112     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 113     __m128i          ewitab;
 114     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 115     real             *ewtab;
 116     __m128d          dummy_mask,cutoff_mask;
 117     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 118     __m128d          one     = _mm_set1_pd(1.0);
 119     __m128d          two     = _mm_set1_pd(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm_set1_pd(fr->ic->epsfac);
 132     charge           = mdatoms->chargeA;
 133     nvdwtype         = fr->ntype;
 134     vdwparam         = fr->nbfp;
 135     vdwtype          = mdatoms->typeA;
 136
 137     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 138     ewtab            = fr->ic->tabq_coul_FDV0;
 139     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 140     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 145     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 146     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 147     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149     jq1              = _mm_set1_pd(charge[inr+1]);
 150     jq2              = _mm_set1_pd(charge[inr+2]);
 151     jq3              = _mm_set1_pd(charge[inr+3]);
 152     vdwjidx0A        = 2*vdwtype[inr+0];
 153     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 154     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 155     qq11             = _mm_mul_pd(iq1,jq1);
 156     qq12             = _mm_mul_pd(iq1,jq2);
 157     qq13             = _mm_mul_pd(iq1,jq3);
 158     qq21             = _mm_mul_pd(iq2,jq1);
 159     qq22             = _mm_mul_pd(iq2,jq2);
 160     qq23             = _mm_mul_pd(iq2,jq3);
 161     qq31             = _mm_mul_pd(iq3,jq1);
 162     qq32             = _mm_mul_pd(iq3,jq2);
 163     qq33             = _mm_mul_pd(iq3,jq3);
 164
 165     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 166     rcutoff_scalar   = fr->ic->rcoulomb;
 167     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 168     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 169
 170     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 171     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 172
 173     /* Avoid stupid compiler warnings */
 174     jnrA = jnrB = 0;
 175     j_coord_offsetA = 0;
 176     j_coord_offsetB = 0;
 177
 178     outeriter        = 0;
 179     inneriter        = 0;
 180
 181     /* Start outer loop over neighborlists */
 182     for(iidx=0; iidx<nri; iidx++)
 183     {
 184         /* Load shift vector for this list */
 185         i_shift_offset   = DIM*shiftidx[iidx];
 186
 187         /* Load limits for loop over neighbors */
 188         j_index_start    = jindex[iidx];
 189         j_index_end      = jindex[iidx+1];
 190
 191         /* Get outer coordinate index */
 192         inr              = iinr[iidx];
 193         i_coord_offset   = DIM*inr;
 194
 195         /* Load i particle coords and add shift vector */
 196         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 197                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 198
 199         fix0             = _mm_setzero_pd();
 200         fiy0             = _mm_setzero_pd();
 201         fiz0             = _mm_setzero_pd();
 202         fix1             = _mm_setzero_pd();
 203         fiy1             = _mm_setzero_pd();
 204         fiz1             = _mm_setzero_pd();
 205         fix2             = _mm_setzero_pd();
 206         fiy2             = _mm_setzero_pd();
 207         fiz2             = _mm_setzero_pd();
 208         fix3             = _mm_setzero_pd();
 209         fiy3             = _mm_setzero_pd();
 210         fiz3             = _mm_setzero_pd();
 211
 212         /* Reset potential sums */
 213         velecsum         = _mm_setzero_pd();
 214         vvdwsum          = _mm_setzero_pd();
 215
 216         /* Start inner kernel loop */
 217         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 218         {
 219
 220             /* Get j neighbor index, and coordinate index */
 221             jnrA             = jjnr[jidx];
 222             jnrB             = jjnr[jidx+1];
 223             j_coord_offsetA  = DIM*jnrA;
 224             j_coord_offsetB  = DIM*jnrB;
 225
 226             /* load j atom coordinates */
 227             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 228                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 229                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 230
 231             /* Calculate displacement vector */
 232             dx00             = _mm_sub_pd(ix0,jx0);
 233             dy00             = _mm_sub_pd(iy0,jy0);
 234             dz00             = _mm_sub_pd(iz0,jz0);
 235             dx11             = _mm_sub_pd(ix1,jx1);
 236             dy11             = _mm_sub_pd(iy1,jy1);
 237             dz11             = _mm_sub_pd(iz1,jz1);
 238             dx12             = _mm_sub_pd(ix1,jx2);
 239             dy12             = _mm_sub_pd(iy1,jy2);
 240             dz12             = _mm_sub_pd(iz1,jz2);
 241             dx13             = _mm_sub_pd(ix1,jx3);
 242             dy13             = _mm_sub_pd(iy1,jy3);
 243             dz13             = _mm_sub_pd(iz1,jz3);
 244             dx21             = _mm_sub_pd(ix2,jx1);
 245             dy21             = _mm_sub_pd(iy2,jy1);
 246             dz21             = _mm_sub_pd(iz2,jz1);
 247             dx22             = _mm_sub_pd(ix2,jx2);
 248             dy22             = _mm_sub_pd(iy2,jy2);
 249             dz22             = _mm_sub_pd(iz2,jz2);
 250             dx23             = _mm_sub_pd(ix2,jx3);
 251             dy23             = _mm_sub_pd(iy2,jy3);
 252             dz23             = _mm_sub_pd(iz2,jz3);
 253             dx31             = _mm_sub_pd(ix3,jx1);
 254             dy31             = _mm_sub_pd(iy3,jy1);
 255             dz31             = _mm_sub_pd(iz3,jz1);
 256             dx32             = _mm_sub_pd(ix3,jx2);
 257             dy32             = _mm_sub_pd(iy3,jy2);
 258             dz32             = _mm_sub_pd(iz3,jz2);
 259             dx33             = _mm_sub_pd(ix3,jx3);
 260             dy33             = _mm_sub_pd(iy3,jy3);
 261             dz33             = _mm_sub_pd(iz3,jz3);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 265             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 266             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 267             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 268             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 269             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 270             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 271             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 272             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 273             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 274
 275             rinv11           = avx128fma_invsqrt_d(rsq11);
 276             rinv12           = avx128fma_invsqrt_d(rsq12);
 277             rinv13           = avx128fma_invsqrt_d(rsq13);
 278             rinv21           = avx128fma_invsqrt_d(rsq21);
 279             rinv22           = avx128fma_invsqrt_d(rsq22);
 280             rinv23           = avx128fma_invsqrt_d(rsq23);
 281             rinv31           = avx128fma_invsqrt_d(rsq31);
 282             rinv32           = avx128fma_invsqrt_d(rsq32);
 283             rinv33           = avx128fma_invsqrt_d(rsq33);
 284
 285             rinvsq00         = avx128fma_inv_d(rsq00);
 286             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 287             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 288             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 289             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 290             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 291             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 292             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 293             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 294             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 295
 296             fjx0             = _mm_setzero_pd();
 297             fjy0             = _mm_setzero_pd();
 298             fjz0             = _mm_setzero_pd();
 299             fjx1             = _mm_setzero_pd();
 300             fjy1             = _mm_setzero_pd();
 301             fjz1             = _mm_setzero_pd();
 302             fjx2             = _mm_setzero_pd();
 303             fjy2             = _mm_setzero_pd();
 304             fjz2             = _mm_setzero_pd();
 305             fjx3             = _mm_setzero_pd();
 306             fjy3             = _mm_setzero_pd();
 307             fjz3             = _mm_setzero_pd();
 308
 309             /**************************
 310              * CALCULATE INTERACTIONS *
 311              **************************/
 312
 313             if (gmx_mm_any_lt(rsq00,rcutoff2))
 314             {
 315
 316             /* LENNARD-JONES DISPERSION/REPULSION */
 317
 318             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 319             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 320             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 321             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 322                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 323             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 324
 325             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 329             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 330
 331             fscal            = fvdw;
 332
 333             fscal            = _mm_and_pd(fscal,cutoff_mask);
 334
 335             /* Update vectorial force */
 336             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 337             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 338             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 339
 340             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 341             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 342             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 343
 344             }
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             if (gmx_mm_any_lt(rsq11,rcutoff2))
 351             {
 352
 353             r11              = _mm_mul_pd(rsq11,rinv11);
 354
 355             /* EWALD ELECTROSTATICS */
 356
 357             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 358             ewrt             = _mm_mul_pd(r11,ewtabscale);
 359             ewitab           = _mm_cvttpd_epi32(ewrt);
 360 #ifdef __XOP__
 361             eweps            = _mm_frcz_pd(ewrt);
 362 #else
 363             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 364 #endif
 365             twoeweps         = _mm_add_pd(eweps,eweps);
 366             ewitab           = _mm_slli_epi32(ewitab,2);
 367             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 368             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 369             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 370             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 371             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 372             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 373             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 374             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 375             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
 376             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 377
 378             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velec            = _mm_and_pd(velec,cutoff_mask);
 382             velecsum         = _mm_add_pd(velecsum,velec);
 383
 384             fscal            = felec;
 385
 386             fscal            = _mm_and_pd(fscal,cutoff_mask);
 387
 388             /* Update vectorial force */
 389             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 390             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 391             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 392
 393             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 394             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 395             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 396
 397             }
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             if (gmx_mm_any_lt(rsq12,rcutoff2))
 404             {
 405
 406             r12              = _mm_mul_pd(rsq12,rinv12);
 407
 408             /* EWALD ELECTROSTATICS */
 409
 410             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 411             ewrt             = _mm_mul_pd(r12,ewtabscale);
 412             ewitab           = _mm_cvttpd_epi32(ewrt);
 413 #ifdef __XOP__
 414             eweps            = _mm_frcz_pd(ewrt);
 415 #else
 416             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 417 #endif
 418             twoeweps         = _mm_add_pd(eweps,eweps);
 419             ewitab           = _mm_slli_epi32(ewitab,2);
 420             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 421             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 422             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 423             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 424             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 425             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 426             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 427             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 428             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
 429             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 430
 431             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velec            = _mm_and_pd(velec,cutoff_mask);
 435             velecsum         = _mm_add_pd(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             fscal            = _mm_and_pd(fscal,cutoff_mask);
 440
 441             /* Update vectorial force */
 442             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 443             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 444             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 445
 446             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 447             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 448             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 449
 450             }
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             if (gmx_mm_any_lt(rsq13,rcutoff2))
 457             {
 458
 459             r13              = _mm_mul_pd(rsq13,rinv13);
 460
 461             /* EWALD ELECTROSTATICS */
 462
 463             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 464             ewrt             = _mm_mul_pd(r13,ewtabscale);
 465             ewitab           = _mm_cvttpd_epi32(ewrt);
 466 #ifdef __XOP__
 467             eweps            = _mm_frcz_pd(ewrt);
 468 #else
 469             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 470 #endif
 471             twoeweps         = _mm_add_pd(eweps,eweps);
 472             ewitab           = _mm_slli_epi32(ewitab,2);
 473             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 474             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 475             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 476             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 477             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 478             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 479             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 480             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 481             velec            = _mm_mul_pd(qq13,_mm_sub_pd(_mm_sub_pd(rinv13,sh_ewald),velec));
 482             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 483
 484             cutoff_mask      = _mm_cmplt_pd(rsq13,rcutoff2);
 485
 486             /* Update potential sum for this i atom from the interaction with this j atom. */
 487             velec            = _mm_and_pd(velec,cutoff_mask);
 488             velecsum         = _mm_add_pd(velecsum,velec);
 489
 490             fscal            = felec;
 491
 492             fscal            = _mm_and_pd(fscal,cutoff_mask);
 493
 494             /* Update vectorial force */
 495             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 496             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 497             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 498
 499             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 500             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 501             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 502
 503             }
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             if (gmx_mm_any_lt(rsq21,rcutoff2))
 510             {
 511
 512             r21              = _mm_mul_pd(rsq21,rinv21);
 513
 514             /* EWALD ELECTROSTATICS */
 515
 516             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 517             ewrt             = _mm_mul_pd(r21,ewtabscale);
 518             ewitab           = _mm_cvttpd_epi32(ewrt);
 519 #ifdef __XOP__
 520             eweps            = _mm_frcz_pd(ewrt);
 521 #else
 522             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 523 #endif
 524             twoeweps         = _mm_add_pd(eweps,eweps);
 525             ewitab           = _mm_slli_epi32(ewitab,2);
 526             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 527             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 528             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 529             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 530             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 531             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 532             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 533             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 534             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
 535             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 536
 537             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
 538
 539             /* Update potential sum for this i atom from the interaction with this j atom. */
 540             velec            = _mm_and_pd(velec,cutoff_mask);
 541             velecsum         = _mm_add_pd(velecsum,velec);
 542
 543             fscal            = felec;
 544
 545             fscal            = _mm_and_pd(fscal,cutoff_mask);
 546
 547             /* Update vectorial force */
 548             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 549             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 550             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 551
 552             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 553             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 554             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 555
 556             }
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             if (gmx_mm_any_lt(rsq22,rcutoff2))
 563             {
 564
 565             r22              = _mm_mul_pd(rsq22,rinv22);
 566
 567             /* EWALD ELECTROSTATICS */
 568
 569             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 570             ewrt             = _mm_mul_pd(r22,ewtabscale);
 571             ewitab           = _mm_cvttpd_epi32(ewrt);
 572 #ifdef __XOP__
 573             eweps            = _mm_frcz_pd(ewrt);
 574 #else
 575             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 576 #endif
 577             twoeweps         = _mm_add_pd(eweps,eweps);
 578             ewitab           = _mm_slli_epi32(ewitab,2);
 579             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 580             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 581             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 582             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 583             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 584             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 585             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 586             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 587             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
 588             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 589
 590             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
 591
 592             /* Update potential sum for this i atom from the interaction with this j atom. */
 593             velec            = _mm_and_pd(velec,cutoff_mask);
 594             velecsum         = _mm_add_pd(velecsum,velec);
 595
 596             fscal            = felec;
 597
 598             fscal            = _mm_and_pd(fscal,cutoff_mask);
 599
 600             /* Update vectorial force */
 601             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 602             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 603             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 604
 605             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 606             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 607             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 608
 609             }
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             if (gmx_mm_any_lt(rsq23,rcutoff2))
 616             {
 617
 618             r23              = _mm_mul_pd(rsq23,rinv23);
 619
 620             /* EWALD ELECTROSTATICS */
 621
 622             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 623             ewrt             = _mm_mul_pd(r23,ewtabscale);
 624             ewitab           = _mm_cvttpd_epi32(ewrt);
 625 #ifdef __XOP__
 626             eweps            = _mm_frcz_pd(ewrt);
 627 #else
 628             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 629 #endif
 630             twoeweps         = _mm_add_pd(eweps,eweps);
 631             ewitab           = _mm_slli_epi32(ewitab,2);
 632             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 633             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 634             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 635             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 636             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 637             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 638             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 639             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 640             velec            = _mm_mul_pd(qq23,_mm_sub_pd(_mm_sub_pd(rinv23,sh_ewald),velec));
 641             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
 642
 643             cutoff_mask      = _mm_cmplt_pd(rsq23,rcutoff2);
 644
 645             /* Update potential sum for this i atom from the interaction with this j atom. */
 646             velec            = _mm_and_pd(velec,cutoff_mask);
 647             velecsum         = _mm_add_pd(velecsum,velec);
 648
 649             fscal            = felec;
 650
 651             fscal            = _mm_and_pd(fscal,cutoff_mask);
 652
 653             /* Update vectorial force */
 654             fix2             = _mm_macc_pd(dx23,fscal,fix2);
 655             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
 656             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
 657
 658             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
 659             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
 660             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
 661
 662             }
 663
 664             /**************************
 665              * CALCULATE INTERACTIONS *
 666              **************************/
 667
 668             if (gmx_mm_any_lt(rsq31,rcutoff2))
 669             {
 670
 671             r31              = _mm_mul_pd(rsq31,rinv31);
 672
 673             /* EWALD ELECTROSTATICS */
 674
 675             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 676             ewrt             = _mm_mul_pd(r31,ewtabscale);
 677             ewitab           = _mm_cvttpd_epi32(ewrt);
 678 #ifdef __XOP__
 679             eweps            = _mm_frcz_pd(ewrt);
 680 #else
 681             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 682 #endif
 683             twoeweps         = _mm_add_pd(eweps,eweps);
 684             ewitab           = _mm_slli_epi32(ewitab,2);
 685             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 686             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 687             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 688             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 689             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 690             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 691             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 692             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 693             velec            = _mm_mul_pd(qq31,_mm_sub_pd(_mm_sub_pd(rinv31,sh_ewald),velec));
 694             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
 695
 696             cutoff_mask      = _mm_cmplt_pd(rsq31,rcutoff2);
 697
 698             /* Update potential sum for this i atom from the interaction with this j atom. */
 699             velec            = _mm_and_pd(velec,cutoff_mask);
 700             velecsum         = _mm_add_pd(velecsum,velec);
 701
 702             fscal            = felec;
 703
 704             fscal            = _mm_and_pd(fscal,cutoff_mask);
 705
 706             /* Update vectorial force */
 707             fix3             = _mm_macc_pd(dx31,fscal,fix3);
 708             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
 709             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
 710
 711             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
 712             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
 713             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
 714
 715             }
 716
 717             /**************************
 718              * CALCULATE INTERACTIONS *
 719              **************************/
 720
 721             if (gmx_mm_any_lt(rsq32,rcutoff2))
 722             {
 723
 724             r32              = _mm_mul_pd(rsq32,rinv32);
 725
 726             /* EWALD ELECTROSTATICS */
 727
 728             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 729             ewrt             = _mm_mul_pd(r32,ewtabscale);
 730             ewitab           = _mm_cvttpd_epi32(ewrt);
 731 #ifdef __XOP__
 732             eweps            = _mm_frcz_pd(ewrt);
 733 #else
 734             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 735 #endif
 736             twoeweps         = _mm_add_pd(eweps,eweps);
 737             ewitab           = _mm_slli_epi32(ewitab,2);
 738             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 739             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 740             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 741             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 742             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 743             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 744             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 745             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 746             velec            = _mm_mul_pd(qq32,_mm_sub_pd(_mm_sub_pd(rinv32,sh_ewald),velec));
 747             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
 748
 749             cutoff_mask      = _mm_cmplt_pd(rsq32,rcutoff2);
 750
 751             /* Update potential sum for this i atom from the interaction with this j atom. */
 752             velec            = _mm_and_pd(velec,cutoff_mask);
 753             velecsum         = _mm_add_pd(velecsum,velec);
 754
 755             fscal            = felec;
 756
 757             fscal            = _mm_and_pd(fscal,cutoff_mask);
 758
 759             /* Update vectorial force */
 760             fix3             = _mm_macc_pd(dx32,fscal,fix3);
 761             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
 762             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
 763
 764             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
 765             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
 766             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
 767
 768             }
 769
 770             /**************************
 771              * CALCULATE INTERACTIONS *
 772              **************************/
 773
 774             if (gmx_mm_any_lt(rsq33,rcutoff2))
 775             {
 776
 777             r33              = _mm_mul_pd(rsq33,rinv33);
 778
 779             /* EWALD ELECTROSTATICS */
 780
 781             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 782             ewrt             = _mm_mul_pd(r33,ewtabscale);
 783             ewitab           = _mm_cvttpd_epi32(ewrt);
 784 #ifdef __XOP__
 785             eweps            = _mm_frcz_pd(ewrt);
 786 #else
 787             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 788 #endif
 789             twoeweps         = _mm_add_pd(eweps,eweps);
 790             ewitab           = _mm_slli_epi32(ewitab,2);
 791             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 792             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 793             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 794             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 795             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 796             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 797             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 798             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 799             velec            = _mm_mul_pd(qq33,_mm_sub_pd(_mm_sub_pd(rinv33,sh_ewald),velec));
 800             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
 801
 802             cutoff_mask      = _mm_cmplt_pd(rsq33,rcutoff2);
 803
 804             /* Update potential sum for this i atom from the interaction with this j atom. */
 805             velec            = _mm_and_pd(velec,cutoff_mask);
 806             velecsum         = _mm_add_pd(velecsum,velec);
 807
 808             fscal            = felec;
 809
 810             fscal            = _mm_and_pd(fscal,cutoff_mask);
 811
 812             /* Update vectorial force */
 813             fix3             = _mm_macc_pd(dx33,fscal,fix3);
 814             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
 815             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
 816
 817             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
 818             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
 819             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
 820
 821             }
 822
 823             gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 824
 825             /* Inner loop uses 488 flops */
 826         }
 827
 828         if(jidx<j_index_end)
 829         {
 830
 831             jnrA             = jjnr[jidx];
 832             j_coord_offsetA  = DIM*jnrA;
 833
 834             /* load j atom coordinates */
 835             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 836                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 837                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 838
 839             /* Calculate displacement vector */
 840             dx00             = _mm_sub_pd(ix0,jx0);
 841             dy00             = _mm_sub_pd(iy0,jy0);
 842             dz00             = _mm_sub_pd(iz0,jz0);
 843             dx11             = _mm_sub_pd(ix1,jx1);
 844             dy11             = _mm_sub_pd(iy1,jy1);
 845             dz11             = _mm_sub_pd(iz1,jz1);
 846             dx12             = _mm_sub_pd(ix1,jx2);
 847             dy12             = _mm_sub_pd(iy1,jy2);
 848             dz12             = _mm_sub_pd(iz1,jz2);
 849             dx13             = _mm_sub_pd(ix1,jx3);
 850             dy13             = _mm_sub_pd(iy1,jy3);
 851             dz13             = _mm_sub_pd(iz1,jz3);
 852             dx21             = _mm_sub_pd(ix2,jx1);
 853             dy21             = _mm_sub_pd(iy2,jy1);
 854             dz21             = _mm_sub_pd(iz2,jz1);
 855             dx22             = _mm_sub_pd(ix2,jx2);
 856             dy22             = _mm_sub_pd(iy2,jy2);
 857             dz22             = _mm_sub_pd(iz2,jz2);
 858             dx23             = _mm_sub_pd(ix2,jx3);
 859             dy23             = _mm_sub_pd(iy2,jy3);
 860             dz23             = _mm_sub_pd(iz2,jz3);
 861             dx31             = _mm_sub_pd(ix3,jx1);
 862             dy31             = _mm_sub_pd(iy3,jy1);
 863             dz31             = _mm_sub_pd(iz3,jz1);
 864             dx32             = _mm_sub_pd(ix3,jx2);
 865             dy32             = _mm_sub_pd(iy3,jy2);
 866             dz32             = _mm_sub_pd(iz3,jz2);
 867             dx33             = _mm_sub_pd(ix3,jx3);
 868             dy33             = _mm_sub_pd(iy3,jy3);
 869             dz33             = _mm_sub_pd(iz3,jz3);
 870
 871             /* Calculate squared distance and things based on it */
 872             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 873             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 874             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 875             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 876             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 877             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 878             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 879             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 880             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 881             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 882
 883             rinv11           = avx128fma_invsqrt_d(rsq11);
 884             rinv12           = avx128fma_invsqrt_d(rsq12);
 885             rinv13           = avx128fma_invsqrt_d(rsq13);
 886             rinv21           = avx128fma_invsqrt_d(rsq21);
 887             rinv22           = avx128fma_invsqrt_d(rsq22);
 888             rinv23           = avx128fma_invsqrt_d(rsq23);
 889             rinv31           = avx128fma_invsqrt_d(rsq31);
 890             rinv32           = avx128fma_invsqrt_d(rsq32);
 891             rinv33           = avx128fma_invsqrt_d(rsq33);
 892
 893             rinvsq00         = avx128fma_inv_d(rsq00);
 894             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 895             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 896             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 897             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 898             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 899             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 900             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 901             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 902             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 903
 904             fjx0             = _mm_setzero_pd();
 905             fjy0             = _mm_setzero_pd();
 906             fjz0             = _mm_setzero_pd();
 907             fjx1             = _mm_setzero_pd();
 908             fjy1             = _mm_setzero_pd();
 909             fjz1             = _mm_setzero_pd();
 910             fjx2             = _mm_setzero_pd();
 911             fjy2             = _mm_setzero_pd();
 912             fjz2             = _mm_setzero_pd();
 913             fjx3             = _mm_setzero_pd();
 914             fjy3             = _mm_setzero_pd();
 915             fjz3             = _mm_setzero_pd();
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             if (gmx_mm_any_lt(rsq00,rcutoff2))
 922             {
 923
 924             /* LENNARD-JONES DISPERSION/REPULSION */
 925
 926             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 927             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 928             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 929             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 930                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 931             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 932
 933             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 934
 935             /* Update potential sum for this i atom from the interaction with this j atom. */
 936             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 937             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 938             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 939
 940             fscal            = fvdw;
 941
 942             fscal            = _mm_and_pd(fscal,cutoff_mask);
 943
 944             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 945
 946             /* Update vectorial force */
 947             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 948             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 949             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 950
 951             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 952             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 953             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 954
 955             }
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             if (gmx_mm_any_lt(rsq11,rcutoff2))
 962             {
 963
 964             r11              = _mm_mul_pd(rsq11,rinv11);
 965
 966             /* EWALD ELECTROSTATICS */
 967
 968             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 969             ewrt             = _mm_mul_pd(r11,ewtabscale);
 970             ewitab           = _mm_cvttpd_epi32(ewrt);
 971 #ifdef __XOP__
 972             eweps            = _mm_frcz_pd(ewrt);
 973 #else
 974             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 975 #endif
 976             twoeweps         = _mm_add_pd(eweps,eweps);
 977             ewitab           = _mm_slli_epi32(ewitab,2);
 978             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 979             ewtabD           = _mm_setzero_pd();
 980             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 981             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 982             ewtabFn          = _mm_setzero_pd();
 983             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 984             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 985             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 986             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
 987             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 988
 989             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
 990
 991             /* Update potential sum for this i atom from the interaction with this j atom. */
 992             velec            = _mm_and_pd(velec,cutoff_mask);
 993             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 994             velecsum         = _mm_add_pd(velecsum,velec);
 995
 996             fscal            = felec;
 997
 998             fscal            = _mm_and_pd(fscal,cutoff_mask);
 999
1000             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1001
1002             /* Update vectorial force */
1003             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1004             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1005             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1006
1007             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1008             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1009             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1010
1011             }
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             if (gmx_mm_any_lt(rsq12,rcutoff2))
1018             {
1019
1020             r12              = _mm_mul_pd(rsq12,rinv12);
1021
1022             /* EWALD ELECTROSTATICS */
1023
1024             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1025             ewrt             = _mm_mul_pd(r12,ewtabscale);
1026             ewitab           = _mm_cvttpd_epi32(ewrt);
1027 #ifdef __XOP__
1028             eweps            = _mm_frcz_pd(ewrt);
1029 #else
1030             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1031 #endif
1032             twoeweps         = _mm_add_pd(eweps,eweps);
1033             ewitab           = _mm_slli_epi32(ewitab,2);
1034             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1035             ewtabD           = _mm_setzero_pd();
1036             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1037             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1038             ewtabFn          = _mm_setzero_pd();
1039             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1040             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1041             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1042             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
1043             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1044
1045             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
1046
1047             /* Update potential sum for this i atom from the interaction with this j atom. */
1048             velec            = _mm_and_pd(velec,cutoff_mask);
1049             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1050             velecsum         = _mm_add_pd(velecsum,velec);
1051
1052             fscal            = felec;
1053
1054             fscal            = _mm_and_pd(fscal,cutoff_mask);
1055
1056             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1057
1058             /* Update vectorial force */
1059             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1060             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1061             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1062
1063             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1064             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1065             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1066
1067             }
1068
1069             /**************************
1070              * CALCULATE INTERACTIONS *
1071              **************************/
1072
1073             if (gmx_mm_any_lt(rsq13,rcutoff2))
1074             {
1075
1076             r13              = _mm_mul_pd(rsq13,rinv13);
1077
1078             /* EWALD ELECTROSTATICS */
1079
1080             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1081             ewrt             = _mm_mul_pd(r13,ewtabscale);
1082             ewitab           = _mm_cvttpd_epi32(ewrt);
1083 #ifdef __XOP__
1084             eweps            = _mm_frcz_pd(ewrt);
1085 #else
1086             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1087 #endif
1088             twoeweps         = _mm_add_pd(eweps,eweps);
1089             ewitab           = _mm_slli_epi32(ewitab,2);
1090             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1091             ewtabD           = _mm_setzero_pd();
1092             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1093             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1094             ewtabFn          = _mm_setzero_pd();
1095             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1096             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1097             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1098             velec            = _mm_mul_pd(qq13,_mm_sub_pd(_mm_sub_pd(rinv13,sh_ewald),velec));
1099             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1100
1101             cutoff_mask      = _mm_cmplt_pd(rsq13,rcutoff2);
1102
1103             /* Update potential sum for this i atom from the interaction with this j atom. */
1104             velec            = _mm_and_pd(velec,cutoff_mask);
1105             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1106             velecsum         = _mm_add_pd(velecsum,velec);
1107
1108             fscal            = felec;
1109
1110             fscal            = _mm_and_pd(fscal,cutoff_mask);
1111
1112             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1113
1114             /* Update vectorial force */
1115             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1116             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1117             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1118
1119             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1120             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1121             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1122
1123             }
1124
1125             /**************************
1126              * CALCULATE INTERACTIONS *
1127              **************************/
1128
1129             if (gmx_mm_any_lt(rsq21,rcutoff2))
1130             {
1131
1132             r21              = _mm_mul_pd(rsq21,rinv21);
1133
1134             /* EWALD ELECTROSTATICS */
1135
1136             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1137             ewrt             = _mm_mul_pd(r21,ewtabscale);
1138             ewitab           = _mm_cvttpd_epi32(ewrt);
1139 #ifdef __XOP__
1140             eweps            = _mm_frcz_pd(ewrt);
1141 #else
1142             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1143 #endif
1144             twoeweps         = _mm_add_pd(eweps,eweps);
1145             ewitab           = _mm_slli_epi32(ewitab,2);
1146             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1147             ewtabD           = _mm_setzero_pd();
1148             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1149             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1150             ewtabFn          = _mm_setzero_pd();
1151             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1152             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1153             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1154             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
1155             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1156
1157             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
1158
1159             /* Update potential sum for this i atom from the interaction with this j atom. */
1160             velec            = _mm_and_pd(velec,cutoff_mask);
1161             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1162             velecsum         = _mm_add_pd(velecsum,velec);
1163
1164             fscal            = felec;
1165
1166             fscal            = _mm_and_pd(fscal,cutoff_mask);
1167
1168             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1169
1170             /* Update vectorial force */
1171             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1172             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1173             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1174
1175             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1176             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1177             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1178
1179             }
1180
1181             /**************************
1182              * CALCULATE INTERACTIONS *
1183              **************************/
1184
1185             if (gmx_mm_any_lt(rsq22,rcutoff2))
1186             {
1187
1188             r22              = _mm_mul_pd(rsq22,rinv22);
1189
1190             /* EWALD ELECTROSTATICS */
1191
1192             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1193             ewrt             = _mm_mul_pd(r22,ewtabscale);
1194             ewitab           = _mm_cvttpd_epi32(ewrt);
1195 #ifdef __XOP__
1196             eweps            = _mm_frcz_pd(ewrt);
1197 #else
1198             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1199 #endif
1200             twoeweps         = _mm_add_pd(eweps,eweps);
1201             ewitab           = _mm_slli_epi32(ewitab,2);
1202             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1203             ewtabD           = _mm_setzero_pd();
1204             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1205             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1206             ewtabFn          = _mm_setzero_pd();
1207             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1208             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1209             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1210             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
1211             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1212
1213             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
1214
1215             /* Update potential sum for this i atom from the interaction with this j atom. */
1216             velec            = _mm_and_pd(velec,cutoff_mask);
1217             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1218             velecsum         = _mm_add_pd(velecsum,velec);
1219
1220             fscal            = felec;
1221
1222             fscal            = _mm_and_pd(fscal,cutoff_mask);
1223
1224             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1225
1226             /* Update vectorial force */
1227             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1228             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1229             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1230
1231             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1232             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1233             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1234
1235             }
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             if (gmx_mm_any_lt(rsq23,rcutoff2))
1242             {
1243
1244             r23              = _mm_mul_pd(rsq23,rinv23);
1245
1246             /* EWALD ELECTROSTATICS */
1247
1248             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1249             ewrt             = _mm_mul_pd(r23,ewtabscale);
1250             ewitab           = _mm_cvttpd_epi32(ewrt);
1251 #ifdef __XOP__
1252             eweps            = _mm_frcz_pd(ewrt);
1253 #else
1254             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1255 #endif
1256             twoeweps         = _mm_add_pd(eweps,eweps);
1257             ewitab           = _mm_slli_epi32(ewitab,2);
1258             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1259             ewtabD           = _mm_setzero_pd();
1260             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1261             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1262             ewtabFn          = _mm_setzero_pd();
1263             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1264             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1265             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1266             velec            = _mm_mul_pd(qq23,_mm_sub_pd(_mm_sub_pd(rinv23,sh_ewald),velec));
1267             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1268
1269             cutoff_mask      = _mm_cmplt_pd(rsq23,rcutoff2);
1270
1271             /* Update potential sum for this i atom from the interaction with this j atom. */
1272             velec            = _mm_and_pd(velec,cutoff_mask);
1273             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1274             velecsum         = _mm_add_pd(velecsum,velec);
1275
1276             fscal            = felec;
1277
1278             fscal            = _mm_and_pd(fscal,cutoff_mask);
1279
1280             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1281
1282             /* Update vectorial force */
1283             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1284             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1285             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1286
1287             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1288             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1289             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1290
1291             }
1292
1293             /**************************
1294              * CALCULATE INTERACTIONS *
1295              **************************/
1296
1297             if (gmx_mm_any_lt(rsq31,rcutoff2))
1298             {
1299
1300             r31              = _mm_mul_pd(rsq31,rinv31);
1301
1302             /* EWALD ELECTROSTATICS */
1303
1304             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1305             ewrt             = _mm_mul_pd(r31,ewtabscale);
1306             ewitab           = _mm_cvttpd_epi32(ewrt);
1307 #ifdef __XOP__
1308             eweps            = _mm_frcz_pd(ewrt);
1309 #else
1310             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1311 #endif
1312             twoeweps         = _mm_add_pd(eweps,eweps);
1313             ewitab           = _mm_slli_epi32(ewitab,2);
1314             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1315             ewtabD           = _mm_setzero_pd();
1316             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1317             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1318             ewtabFn          = _mm_setzero_pd();
1319             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1320             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1321             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1322             velec            = _mm_mul_pd(qq31,_mm_sub_pd(_mm_sub_pd(rinv31,sh_ewald),velec));
1323             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1324
1325             cutoff_mask      = _mm_cmplt_pd(rsq31,rcutoff2);
1326
1327             /* Update potential sum for this i atom from the interaction with this j atom. */
1328             velec            = _mm_and_pd(velec,cutoff_mask);
1329             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1330             velecsum         = _mm_add_pd(velecsum,velec);
1331
1332             fscal            = felec;
1333
1334             fscal            = _mm_and_pd(fscal,cutoff_mask);
1335
1336             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1337
1338             /* Update vectorial force */
1339             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1340             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1341             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1342
1343             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1344             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1345             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1346
1347             }
1348
1349             /**************************
1350              * CALCULATE INTERACTIONS *
1351              **************************/
1352
1353             if (gmx_mm_any_lt(rsq32,rcutoff2))
1354             {
1355
1356             r32              = _mm_mul_pd(rsq32,rinv32);
1357
1358             /* EWALD ELECTROSTATICS */
1359
1360             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1361             ewrt             = _mm_mul_pd(r32,ewtabscale);
1362             ewitab           = _mm_cvttpd_epi32(ewrt);
1363 #ifdef __XOP__
1364             eweps            = _mm_frcz_pd(ewrt);
1365 #else
1366             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1367 #endif
1368             twoeweps         = _mm_add_pd(eweps,eweps);
1369             ewitab           = _mm_slli_epi32(ewitab,2);
1370             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1371             ewtabD           = _mm_setzero_pd();
1372             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1373             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1374             ewtabFn          = _mm_setzero_pd();
1375             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1376             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1377             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1378             velec            = _mm_mul_pd(qq32,_mm_sub_pd(_mm_sub_pd(rinv32,sh_ewald),velec));
1379             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1380
1381             cutoff_mask      = _mm_cmplt_pd(rsq32,rcutoff2);
1382
1383             /* Update potential sum for this i atom from the interaction with this j atom. */
1384             velec            = _mm_and_pd(velec,cutoff_mask);
1385             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1386             velecsum         = _mm_add_pd(velecsum,velec);
1387
1388             fscal            = felec;
1389
1390             fscal            = _mm_and_pd(fscal,cutoff_mask);
1391
1392             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1393
1394             /* Update vectorial force */
1395             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1396             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1397             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1398
1399             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1400             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1401             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1402
1403             }
1404
1405             /**************************
1406              * CALCULATE INTERACTIONS *
1407              **************************/
1408
1409             if (gmx_mm_any_lt(rsq33,rcutoff2))
1410             {
1411
1412             r33              = _mm_mul_pd(rsq33,rinv33);
1413
1414             /* EWALD ELECTROSTATICS */
1415
1416             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1417             ewrt             = _mm_mul_pd(r33,ewtabscale);
1418             ewitab           = _mm_cvttpd_epi32(ewrt);
1419 #ifdef __XOP__
1420             eweps            = _mm_frcz_pd(ewrt);
1421 #else
1422             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1423 #endif
1424             twoeweps         = _mm_add_pd(eweps,eweps);
1425             ewitab           = _mm_slli_epi32(ewitab,2);
1426             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1427             ewtabD           = _mm_setzero_pd();
1428             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1429             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1430             ewtabFn          = _mm_setzero_pd();
1431             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1432             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1433             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1434             velec            = _mm_mul_pd(qq33,_mm_sub_pd(_mm_sub_pd(rinv33,sh_ewald),velec));
1435             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1436
1437             cutoff_mask      = _mm_cmplt_pd(rsq33,rcutoff2);
1438
1439             /* Update potential sum for this i atom from the interaction with this j atom. */
1440             velec            = _mm_and_pd(velec,cutoff_mask);
1441             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1442             velecsum         = _mm_add_pd(velecsum,velec);
1443
1444             fscal            = felec;
1445
1446             fscal            = _mm_and_pd(fscal,cutoff_mask);
1447
1448             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1449
1450             /* Update vectorial force */
1451             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1452             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1453             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1454
1455             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1456             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1457             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1458
1459             }
1460
1461             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1462
1463             /* Inner loop uses 488 flops */
1464         }
1465
1466         /* End of innermost loop */
1467
1468         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1469                                               f+i_coord_offset,fshift+i_shift_offset);
1470
1471         ggid                        = gid[iidx];
1472         /* Update potential energies */
1473         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1474         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1475
1476         /* Increment number of inner iterations */
1477         inneriter                  += j_index_end - j_index_start;
1478
1479         /* Outer loop uses 26 flops */
1480     }
1481
1482     /* Increment number of outer iterations */
1483     outeriter        += nri;
1484
1485     /* Update outer/inner flops */
1486
1487     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*488);
1488 }
1489 /*
1490  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_F_avx_128_fma_double
1491  * Electrostatics interaction: Ewald
1492  * VdW interaction:            LennardJones
1493  * Geometry:                   Water4-Water4
1494  * Calculate force/pot:        Force
1495  */
1496 void
1497 nb_kernel_ElecEwSh_VdwLJSh_GeomW4W4_F_avx_128_fma_double
1498                     (t_nblist                    * gmx_restrict       nlist,
1499                      rvec                        * gmx_restrict          xx,
1500                      rvec                        * gmx_restrict          ff,
1501                      struct t_forcerec           * gmx_restrict          fr,
1502                      t_mdatoms                   * gmx_restrict     mdatoms,
1503                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1504                      t_nrnb                      * gmx_restrict        nrnb)
1505 {
1506     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1507      * just 0 for non-waters.
1508      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1509      * jnr indices corresponding to data put in the four positions in the SIMD register.
1510      */
1511     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1512     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1513     int              jnrA,jnrB;
1514     int              j_coord_offsetA,j_coord_offsetB;
1515     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1516     real             rcutoff_scalar;
1517     real             *shiftvec,*fshift,*x,*f;
1518     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1519     int              vdwioffset0;
1520     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1521     int              vdwioffset1;
1522     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1523     int              vdwioffset2;
1524     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1525     int              vdwioffset3;
1526     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1527     int              vdwjidx0A,vdwjidx0B;
1528     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1529     int              vdwjidx1A,vdwjidx1B;
1530     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1531     int              vdwjidx2A,vdwjidx2B;
1532     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1533     int              vdwjidx3A,vdwjidx3B;
1534     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1535     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1536     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1537     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1538     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1539     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1540     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1541     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1542     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1543     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1544     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1545     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1546     real             *charge;
1547     int              nvdwtype;
1548     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1549     int              *vdwtype;
1550     real             *vdwparam;
1551     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1552     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1553     __m128i          ewitab;
1554     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1555     real             *ewtab;
1556     __m128d          dummy_mask,cutoff_mask;
1557     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1558     __m128d          one     = _mm_set1_pd(1.0);
1559     __m128d          two     = _mm_set1_pd(2.0);
1560     x                = xx[0];
1561     f                = ff[0];
1562
1563     nri              = nlist->nri;
1564     iinr             = nlist->iinr;
1565     jindex           = nlist->jindex;
1566     jjnr             = nlist->jjnr;
1567     shiftidx         = nlist->shift;
1568     gid              = nlist->gid;
1569     shiftvec         = fr->shift_vec[0];
1570     fshift           = fr->fshift[0];
1571     facel            = _mm_set1_pd(fr->ic->epsfac);
1572     charge           = mdatoms->chargeA;
1573     nvdwtype         = fr->ntype;
1574     vdwparam         = fr->nbfp;
1575     vdwtype          = mdatoms->typeA;
1576
1577     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1578     ewtab            = fr->ic->tabq_coul_F;
1579     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1580     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1581
1582     /* Setup water-specific parameters */
1583     inr              = nlist->iinr[0];
1584     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1585     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1586     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1587     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1588
1589     jq1              = _mm_set1_pd(charge[inr+1]);
1590     jq2              = _mm_set1_pd(charge[inr+2]);
1591     jq3              = _mm_set1_pd(charge[inr+3]);
1592     vdwjidx0A        = 2*vdwtype[inr+0];
1593     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1594     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1595     qq11             = _mm_mul_pd(iq1,jq1);
1596     qq12             = _mm_mul_pd(iq1,jq2);
1597     qq13             = _mm_mul_pd(iq1,jq3);
1598     qq21             = _mm_mul_pd(iq2,jq1);
1599     qq22             = _mm_mul_pd(iq2,jq2);
1600     qq23             = _mm_mul_pd(iq2,jq3);
1601     qq31             = _mm_mul_pd(iq3,jq1);
1602     qq32             = _mm_mul_pd(iq3,jq2);
1603     qq33             = _mm_mul_pd(iq3,jq3);
1604
1605     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1606     rcutoff_scalar   = fr->ic->rcoulomb;
1607     rcutoff          = _mm_set1_pd(rcutoff_scalar);
1608     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
1609
1610     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
1611     rvdw             = _mm_set1_pd(fr->ic->rvdw);
1612
1613     /* Avoid stupid compiler warnings */
1614     jnrA = jnrB = 0;
1615     j_coord_offsetA = 0;
1616     j_coord_offsetB = 0;
1617
1618     outeriter        = 0;
1619     inneriter        = 0;
1620
1621     /* Start outer loop over neighborlists */
1622     for(iidx=0; iidx<nri; iidx++)
1623     {
1624         /* Load shift vector for this list */
1625         i_shift_offset   = DIM*shiftidx[iidx];
1626
1627         /* Load limits for loop over neighbors */
1628         j_index_start    = jindex[iidx];
1629         j_index_end      = jindex[iidx+1];
1630
1631         /* Get outer coordinate index */
1632         inr              = iinr[iidx];
1633         i_coord_offset   = DIM*inr;
1634
1635         /* Load i particle coords and add shift vector */
1636         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1637                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1638
1639         fix0             = _mm_setzero_pd();
1640         fiy0             = _mm_setzero_pd();
1641         fiz0             = _mm_setzero_pd();
1642         fix1             = _mm_setzero_pd();
1643         fiy1             = _mm_setzero_pd();
1644         fiz1             = _mm_setzero_pd();
1645         fix2             = _mm_setzero_pd();
1646         fiy2             = _mm_setzero_pd();
1647         fiz2             = _mm_setzero_pd();
1648         fix3             = _mm_setzero_pd();
1649         fiy3             = _mm_setzero_pd();
1650         fiz3             = _mm_setzero_pd();
1651
1652         /* Start inner kernel loop */
1653         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1654         {
1655
1656             /* Get j neighbor index, and coordinate index */
1657             jnrA             = jjnr[jidx];
1658             jnrB             = jjnr[jidx+1];
1659             j_coord_offsetA  = DIM*jnrA;
1660             j_coord_offsetB  = DIM*jnrB;
1661
1662             /* load j atom coordinates */
1663             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1664                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1665                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1666
1667             /* Calculate displacement vector */
1668             dx00             = _mm_sub_pd(ix0,jx0);
1669             dy00             = _mm_sub_pd(iy0,jy0);
1670             dz00             = _mm_sub_pd(iz0,jz0);
1671             dx11             = _mm_sub_pd(ix1,jx1);
1672             dy11             = _mm_sub_pd(iy1,jy1);
1673             dz11             = _mm_sub_pd(iz1,jz1);
1674             dx12             = _mm_sub_pd(ix1,jx2);
1675             dy12             = _mm_sub_pd(iy1,jy2);
1676             dz12             = _mm_sub_pd(iz1,jz2);
1677             dx13             = _mm_sub_pd(ix1,jx3);
1678             dy13             = _mm_sub_pd(iy1,jy3);
1679             dz13             = _mm_sub_pd(iz1,jz3);
1680             dx21             = _mm_sub_pd(ix2,jx1);
1681             dy21             = _mm_sub_pd(iy2,jy1);
1682             dz21             = _mm_sub_pd(iz2,jz1);
1683             dx22             = _mm_sub_pd(ix2,jx2);
1684             dy22             = _mm_sub_pd(iy2,jy2);
1685             dz22             = _mm_sub_pd(iz2,jz2);
1686             dx23             = _mm_sub_pd(ix2,jx3);
1687             dy23             = _mm_sub_pd(iy2,jy3);
1688             dz23             = _mm_sub_pd(iz2,jz3);
1689             dx31             = _mm_sub_pd(ix3,jx1);
1690             dy31             = _mm_sub_pd(iy3,jy1);
1691             dz31             = _mm_sub_pd(iz3,jz1);
1692             dx32             = _mm_sub_pd(ix3,jx2);
1693             dy32             = _mm_sub_pd(iy3,jy2);
1694             dz32             = _mm_sub_pd(iz3,jz2);
1695             dx33             = _mm_sub_pd(ix3,jx3);
1696             dy33             = _mm_sub_pd(iy3,jy3);
1697             dz33             = _mm_sub_pd(iz3,jz3);
1698
1699             /* Calculate squared distance and things based on it */
1700             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1701             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1702             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1703             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1704             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1705             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1706             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1707             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1708             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1709             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1710
1711             rinv11           = avx128fma_invsqrt_d(rsq11);
1712             rinv12           = avx128fma_invsqrt_d(rsq12);
1713             rinv13           = avx128fma_invsqrt_d(rsq13);
1714             rinv21           = avx128fma_invsqrt_d(rsq21);
1715             rinv22           = avx128fma_invsqrt_d(rsq22);
1716             rinv23           = avx128fma_invsqrt_d(rsq23);
1717             rinv31           = avx128fma_invsqrt_d(rsq31);
1718             rinv32           = avx128fma_invsqrt_d(rsq32);
1719             rinv33           = avx128fma_invsqrt_d(rsq33);
1720
1721             rinvsq00         = avx128fma_inv_d(rsq00);
1722             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1723             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1724             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1725             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1726             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1727             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1728             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1729             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1730             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1731
1732             fjx0             = _mm_setzero_pd();
1733             fjy0             = _mm_setzero_pd();
1734             fjz0             = _mm_setzero_pd();
1735             fjx1             = _mm_setzero_pd();
1736             fjy1             = _mm_setzero_pd();
1737             fjz1             = _mm_setzero_pd();
1738             fjx2             = _mm_setzero_pd();
1739             fjy2             = _mm_setzero_pd();
1740             fjz2             = _mm_setzero_pd();
1741             fjx3             = _mm_setzero_pd();
1742             fjy3             = _mm_setzero_pd();
1743             fjz3             = _mm_setzero_pd();
1744
1745             /**************************
1746              * CALCULATE INTERACTIONS *
1747              **************************/
1748
1749             if (gmx_mm_any_lt(rsq00,rcutoff2))
1750             {
1751
1752             /* LENNARD-JONES DISPERSION/REPULSION */
1753
1754             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1755             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1756
1757             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1758
1759             fscal            = fvdw;
1760
1761             fscal            = _mm_and_pd(fscal,cutoff_mask);
1762
1763             /* Update vectorial force */
1764             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1765             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1766             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1767
1768             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1769             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1770             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1771
1772             }
1773
1774             /**************************
1775              * CALCULATE INTERACTIONS *
1776              **************************/
1777
1778             if (gmx_mm_any_lt(rsq11,rcutoff2))
1779             {
1780
1781             r11              = _mm_mul_pd(rsq11,rinv11);
1782
1783             /* EWALD ELECTROSTATICS */
1784
1785             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1786             ewrt             = _mm_mul_pd(r11,ewtabscale);
1787             ewitab           = _mm_cvttpd_epi32(ewrt);
1788 #ifdef __XOP__
1789             eweps            = _mm_frcz_pd(ewrt);
1790 #else
1791             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1792 #endif
1793             twoeweps         = _mm_add_pd(eweps,eweps);
1794             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1795                                          &ewtabF,&ewtabFn);
1796             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1797             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1798
1799             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
1800
1801             fscal            = felec;
1802
1803             fscal            = _mm_and_pd(fscal,cutoff_mask);
1804
1805             /* Update vectorial force */
1806             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1807             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1808             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1809
1810             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1811             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1812             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1813
1814             }
1815
1816             /**************************
1817              * CALCULATE INTERACTIONS *
1818              **************************/
1819
1820             if (gmx_mm_any_lt(rsq12,rcutoff2))
1821             {
1822
1823             r12              = _mm_mul_pd(rsq12,rinv12);
1824
1825             /* EWALD ELECTROSTATICS */
1826
1827             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1828             ewrt             = _mm_mul_pd(r12,ewtabscale);
1829             ewitab           = _mm_cvttpd_epi32(ewrt);
1830 #ifdef __XOP__
1831             eweps            = _mm_frcz_pd(ewrt);
1832 #else
1833             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1834 #endif
1835             twoeweps         = _mm_add_pd(eweps,eweps);
1836             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1837                                          &ewtabF,&ewtabFn);
1838             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1839             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1840
1841             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
1842
1843             fscal            = felec;
1844
1845             fscal            = _mm_and_pd(fscal,cutoff_mask);
1846
1847             /* Update vectorial force */
1848             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1849             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1850             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1851
1852             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1853             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1854             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1855
1856             }
1857
1858             /**************************
1859              * CALCULATE INTERACTIONS *
1860              **************************/
1861
1862             if (gmx_mm_any_lt(rsq13,rcutoff2))
1863             {
1864
1865             r13              = _mm_mul_pd(rsq13,rinv13);
1866
1867             /* EWALD ELECTROSTATICS */
1868
1869             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1870             ewrt             = _mm_mul_pd(r13,ewtabscale);
1871             ewitab           = _mm_cvttpd_epi32(ewrt);
1872 #ifdef __XOP__
1873             eweps            = _mm_frcz_pd(ewrt);
1874 #else
1875             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1876 #endif
1877             twoeweps         = _mm_add_pd(eweps,eweps);
1878             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1879                                          &ewtabF,&ewtabFn);
1880             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1881             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1882
1883             cutoff_mask      = _mm_cmplt_pd(rsq13,rcutoff2);
1884
1885             fscal            = felec;
1886
1887             fscal            = _mm_and_pd(fscal,cutoff_mask);
1888
1889             /* Update vectorial force */
1890             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1891             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1892             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1893
1894             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1895             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1896             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1897
1898             }
1899
1900             /**************************
1901              * CALCULATE INTERACTIONS *
1902              **************************/
1903
1904             if (gmx_mm_any_lt(rsq21,rcutoff2))
1905             {
1906
1907             r21              = _mm_mul_pd(rsq21,rinv21);
1908
1909             /* EWALD ELECTROSTATICS */
1910
1911             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1912             ewrt             = _mm_mul_pd(r21,ewtabscale);
1913             ewitab           = _mm_cvttpd_epi32(ewrt);
1914 #ifdef __XOP__
1915             eweps            = _mm_frcz_pd(ewrt);
1916 #else
1917             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1918 #endif
1919             twoeweps         = _mm_add_pd(eweps,eweps);
1920             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1921                                          &ewtabF,&ewtabFn);
1922             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1923             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1924
1925             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
1926
1927             fscal            = felec;
1928
1929             fscal            = _mm_and_pd(fscal,cutoff_mask);
1930
1931             /* Update vectorial force */
1932             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1933             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1934             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1935
1936             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1937             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1938             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1939
1940             }
1941
1942             /**************************
1943              * CALCULATE INTERACTIONS *
1944              **************************/
1945
1946             if (gmx_mm_any_lt(rsq22,rcutoff2))
1947             {
1948
1949             r22              = _mm_mul_pd(rsq22,rinv22);
1950
1951             /* EWALD ELECTROSTATICS */
1952
1953             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1954             ewrt             = _mm_mul_pd(r22,ewtabscale);
1955             ewitab           = _mm_cvttpd_epi32(ewrt);
1956 #ifdef __XOP__
1957             eweps            = _mm_frcz_pd(ewrt);
1958 #else
1959             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1960 #endif
1961             twoeweps         = _mm_add_pd(eweps,eweps);
1962             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1963                                          &ewtabF,&ewtabFn);
1964             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1965             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1966
1967             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
1968
1969             fscal            = felec;
1970
1971             fscal            = _mm_and_pd(fscal,cutoff_mask);
1972
1973             /* Update vectorial force */
1974             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1975             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1976             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1977
1978             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1979             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1980             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1981
1982             }
1983
1984             /**************************
1985              * CALCULATE INTERACTIONS *
1986              **************************/
1987
1988             if (gmx_mm_any_lt(rsq23,rcutoff2))
1989             {
1990
1991             r23              = _mm_mul_pd(rsq23,rinv23);
1992
1993             /* EWALD ELECTROSTATICS */
1994
1995             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1996             ewrt             = _mm_mul_pd(r23,ewtabscale);
1997             ewitab           = _mm_cvttpd_epi32(ewrt);
1998 #ifdef __XOP__
1999             eweps            = _mm_frcz_pd(ewrt);
2000 #else
2001             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2002 #endif
2003             twoeweps         = _mm_add_pd(eweps,eweps);
2004             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
2005                                          &ewtabF,&ewtabFn);
2006             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2007             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
2008
2009             cutoff_mask      = _mm_cmplt_pd(rsq23,rcutoff2);
2010
2011             fscal            = felec;
2012
2013             fscal            = _mm_and_pd(fscal,cutoff_mask);
2014
2015             /* Update vectorial force */
2016             fix2             = _mm_macc_pd(dx23,fscal,fix2);
2017             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
2018             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
2019
2020             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
2021             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
2022             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
2023
2024             }
2025
2026             /**************************
2027              * CALCULATE INTERACTIONS *
2028              **************************/
2029
2030             if (gmx_mm_any_lt(rsq31,rcutoff2))
2031             {
2032
2033             r31              = _mm_mul_pd(rsq31,rinv31);
2034
2035             /* EWALD ELECTROSTATICS */
2036
2037             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2038             ewrt             = _mm_mul_pd(r31,ewtabscale);
2039             ewitab           = _mm_cvttpd_epi32(ewrt);
2040 #ifdef __XOP__
2041             eweps            = _mm_frcz_pd(ewrt);
2042 #else
2043             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2044 #endif
2045             twoeweps         = _mm_add_pd(eweps,eweps);
2046             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
2047                                          &ewtabF,&ewtabFn);
2048             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2049             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2050
2051             cutoff_mask      = _mm_cmplt_pd(rsq31,rcutoff2);
2052
2053             fscal            = felec;
2054
2055             fscal            = _mm_and_pd(fscal,cutoff_mask);
2056
2057             /* Update vectorial force */
2058             fix3             = _mm_macc_pd(dx31,fscal,fix3);
2059             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
2060             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
2061
2062             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
2063             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
2064             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
2065
2066             }
2067
2068             /**************************
2069              * CALCULATE INTERACTIONS *
2070              **************************/
2071
2072             if (gmx_mm_any_lt(rsq32,rcutoff2))
2073             {
2074
2075             r32              = _mm_mul_pd(rsq32,rinv32);
2076
2077             /* EWALD ELECTROSTATICS */
2078
2079             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2080             ewrt             = _mm_mul_pd(r32,ewtabscale);
2081             ewitab           = _mm_cvttpd_epi32(ewrt);
2082 #ifdef __XOP__
2083             eweps            = _mm_frcz_pd(ewrt);
2084 #else
2085             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2086 #endif
2087             twoeweps         = _mm_add_pd(eweps,eweps);
2088             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
2089                                          &ewtabF,&ewtabFn);
2090             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2091             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2092
2093             cutoff_mask      = _mm_cmplt_pd(rsq32,rcutoff2);
2094
2095             fscal            = felec;
2096
2097             fscal            = _mm_and_pd(fscal,cutoff_mask);
2098
2099             /* Update vectorial force */
2100             fix3             = _mm_macc_pd(dx32,fscal,fix3);
2101             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
2102             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
2103
2104             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
2105             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
2106             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
2107
2108             }
2109
2110             /**************************
2111              * CALCULATE INTERACTIONS *
2112              **************************/
2113
2114             if (gmx_mm_any_lt(rsq33,rcutoff2))
2115             {
2116
2117             r33              = _mm_mul_pd(rsq33,rinv33);
2118
2119             /* EWALD ELECTROSTATICS */
2120
2121             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2122             ewrt             = _mm_mul_pd(r33,ewtabscale);
2123             ewitab           = _mm_cvttpd_epi32(ewrt);
2124 #ifdef __XOP__
2125             eweps            = _mm_frcz_pd(ewrt);
2126 #else
2127             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2128 #endif
2129             twoeweps         = _mm_add_pd(eweps,eweps);
2130             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
2131                                          &ewtabF,&ewtabFn);
2132             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2133             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2134
2135             cutoff_mask      = _mm_cmplt_pd(rsq33,rcutoff2);
2136
2137             fscal            = felec;
2138
2139             fscal            = _mm_and_pd(fscal,cutoff_mask);
2140
2141             /* Update vectorial force */
2142             fix3             = _mm_macc_pd(dx33,fscal,fix3);
2143             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
2144             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
2145
2146             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
2147             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
2148             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
2149
2150             }
2151
2152             gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2153
2154             /* Inner loop uses 414 flops */
2155         }
2156
2157         if(jidx<j_index_end)
2158         {
2159
2160             jnrA             = jjnr[jidx];
2161             j_coord_offsetA  = DIM*jnrA;
2162
2163             /* load j atom coordinates */
2164             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
2165                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2166                                               &jy2,&jz2,&jx3,&jy3,&jz3);
2167
2168             /* Calculate displacement vector */
2169             dx00             = _mm_sub_pd(ix0,jx0);
2170             dy00             = _mm_sub_pd(iy0,jy0);
2171             dz00             = _mm_sub_pd(iz0,jz0);
2172             dx11             = _mm_sub_pd(ix1,jx1);
2173             dy11             = _mm_sub_pd(iy1,jy1);
2174             dz11             = _mm_sub_pd(iz1,jz1);
2175             dx12             = _mm_sub_pd(ix1,jx2);
2176             dy12             = _mm_sub_pd(iy1,jy2);
2177             dz12             = _mm_sub_pd(iz1,jz2);
2178             dx13             = _mm_sub_pd(ix1,jx3);
2179             dy13             = _mm_sub_pd(iy1,jy3);
2180             dz13             = _mm_sub_pd(iz1,jz3);
2181             dx21             = _mm_sub_pd(ix2,jx1);
2182             dy21             = _mm_sub_pd(iy2,jy1);
2183             dz21             = _mm_sub_pd(iz2,jz1);
2184             dx22             = _mm_sub_pd(ix2,jx2);
2185             dy22             = _mm_sub_pd(iy2,jy2);
2186             dz22             = _mm_sub_pd(iz2,jz2);
2187             dx23             = _mm_sub_pd(ix2,jx3);
2188             dy23             = _mm_sub_pd(iy2,jy3);
2189             dz23             = _mm_sub_pd(iz2,jz3);
2190             dx31             = _mm_sub_pd(ix3,jx1);
2191             dy31             = _mm_sub_pd(iy3,jy1);
2192             dz31             = _mm_sub_pd(iz3,jz1);
2193             dx32             = _mm_sub_pd(ix3,jx2);
2194             dy32             = _mm_sub_pd(iy3,jy2);
2195             dz32             = _mm_sub_pd(iz3,jz2);
2196             dx33             = _mm_sub_pd(ix3,jx3);
2197             dy33             = _mm_sub_pd(iy3,jy3);
2198             dz33             = _mm_sub_pd(iz3,jz3);
2199
2200             /* Calculate squared distance and things based on it */
2201             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2202             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2203             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2204             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
2205             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2206             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2207             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
2208             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
2209             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
2210             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
2211
2212             rinv11           = avx128fma_invsqrt_d(rsq11);
2213             rinv12           = avx128fma_invsqrt_d(rsq12);
2214             rinv13           = avx128fma_invsqrt_d(rsq13);
2215             rinv21           = avx128fma_invsqrt_d(rsq21);
2216             rinv22           = avx128fma_invsqrt_d(rsq22);
2217             rinv23           = avx128fma_invsqrt_d(rsq23);
2218             rinv31           = avx128fma_invsqrt_d(rsq31);
2219             rinv32           = avx128fma_invsqrt_d(rsq32);
2220             rinv33           = avx128fma_invsqrt_d(rsq33);
2221
2222             rinvsq00         = avx128fma_inv_d(rsq00);
2223             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
2224             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
2225             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
2226             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
2227             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
2228             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
2229             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
2230             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
2231             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
2232
2233             fjx0             = _mm_setzero_pd();
2234             fjy0             = _mm_setzero_pd();
2235             fjz0             = _mm_setzero_pd();
2236             fjx1             = _mm_setzero_pd();
2237             fjy1             = _mm_setzero_pd();
2238             fjz1             = _mm_setzero_pd();
2239             fjx2             = _mm_setzero_pd();
2240             fjy2             = _mm_setzero_pd();
2241             fjz2             = _mm_setzero_pd();
2242             fjx3             = _mm_setzero_pd();
2243             fjy3             = _mm_setzero_pd();
2244             fjz3             = _mm_setzero_pd();
2245
2246             /**************************
2247              * CALCULATE INTERACTIONS *
2248              **************************/
2249
2250             if (gmx_mm_any_lt(rsq00,rcutoff2))
2251             {
2252
2253             /* LENNARD-JONES DISPERSION/REPULSION */
2254
2255             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2256             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
2257
2258             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
2259
2260             fscal            = fvdw;
2261
2262             fscal            = _mm_and_pd(fscal,cutoff_mask);
2263
2264             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2265
2266             /* Update vectorial force */
2267             fix0             = _mm_macc_pd(dx00,fscal,fix0);
2268             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
2269             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
2270
2271             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
2272             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
2273             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
2274
2275             }
2276
2277             /**************************
2278              * CALCULATE INTERACTIONS *
2279              **************************/
2280
2281             if (gmx_mm_any_lt(rsq11,rcutoff2))
2282             {
2283
2284             r11              = _mm_mul_pd(rsq11,rinv11);
2285
2286             /* EWALD ELECTROSTATICS */
2287
2288             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2289             ewrt             = _mm_mul_pd(r11,ewtabscale);
2290             ewitab           = _mm_cvttpd_epi32(ewrt);
2291 #ifdef __XOP__
2292             eweps            = _mm_frcz_pd(ewrt);
2293 #else
2294             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2295 #endif
2296             twoeweps         = _mm_add_pd(eweps,eweps);
2297             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2298             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2299             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2300
2301             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
2302
2303             fscal            = felec;
2304
2305             fscal            = _mm_and_pd(fscal,cutoff_mask);
2306
2307             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2308
2309             /* Update vectorial force */
2310             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2311             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2312             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2313
2314             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2315             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2316             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2317
2318             }
2319
2320             /**************************
2321              * CALCULATE INTERACTIONS *
2322              **************************/
2323
2324             if (gmx_mm_any_lt(rsq12,rcutoff2))
2325             {
2326
2327             r12              = _mm_mul_pd(rsq12,rinv12);
2328
2329             /* EWALD ELECTROSTATICS */
2330
2331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2332             ewrt             = _mm_mul_pd(r12,ewtabscale);
2333             ewitab           = _mm_cvttpd_epi32(ewrt);
2334 #ifdef __XOP__
2335             eweps            = _mm_frcz_pd(ewrt);
2336 #else
2337             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2338 #endif
2339             twoeweps         = _mm_add_pd(eweps,eweps);
2340             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2341             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2342             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2343
2344             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
2345
2346             fscal            = felec;
2347
2348             fscal            = _mm_and_pd(fscal,cutoff_mask);
2349
2350             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2351
2352             /* Update vectorial force */
2353             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2354             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2355             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2356
2357             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2358             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2359             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2360
2361             }
2362
2363             /**************************
2364              * CALCULATE INTERACTIONS *
2365              **************************/
2366
2367             if (gmx_mm_any_lt(rsq13,rcutoff2))
2368             {
2369
2370             r13              = _mm_mul_pd(rsq13,rinv13);
2371
2372             /* EWALD ELECTROSTATICS */
2373
2374             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2375             ewrt             = _mm_mul_pd(r13,ewtabscale);
2376             ewitab           = _mm_cvttpd_epi32(ewrt);
2377 #ifdef __XOP__
2378             eweps            = _mm_frcz_pd(ewrt);
2379 #else
2380             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2381 #endif
2382             twoeweps         = _mm_add_pd(eweps,eweps);
2383             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2384             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2385             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
2386
2387             cutoff_mask      = _mm_cmplt_pd(rsq13,rcutoff2);
2388
2389             fscal            = felec;
2390
2391             fscal            = _mm_and_pd(fscal,cutoff_mask);
2392
2393             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2394
2395             /* Update vectorial force */
2396             fix1             = _mm_macc_pd(dx13,fscal,fix1);
2397             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
2398             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
2399
2400             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
2401             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
2402             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
2403
2404             }
2405
2406             /**************************
2407              * CALCULATE INTERACTIONS *
2408              **************************/
2409
2410             if (gmx_mm_any_lt(rsq21,rcutoff2))
2411             {
2412
2413             r21              = _mm_mul_pd(rsq21,rinv21);
2414
2415             /* EWALD ELECTROSTATICS */
2416
2417             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2418             ewrt             = _mm_mul_pd(r21,ewtabscale);
2419             ewitab           = _mm_cvttpd_epi32(ewrt);
2420 #ifdef __XOP__
2421             eweps            = _mm_frcz_pd(ewrt);
2422 #else
2423             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2424 #endif
2425             twoeweps         = _mm_add_pd(eweps,eweps);
2426             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2427             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2428             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2429
2430             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
2431
2432             fscal            = felec;
2433
2434             fscal            = _mm_and_pd(fscal,cutoff_mask);
2435
2436             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2437
2438             /* Update vectorial force */
2439             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2440             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2441             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2442
2443             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2444             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2445             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2446
2447             }
2448
2449             /**************************
2450              * CALCULATE INTERACTIONS *
2451              **************************/
2452
2453             if (gmx_mm_any_lt(rsq22,rcutoff2))
2454             {
2455
2456             r22              = _mm_mul_pd(rsq22,rinv22);
2457
2458             /* EWALD ELECTROSTATICS */
2459
2460             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2461             ewrt             = _mm_mul_pd(r22,ewtabscale);
2462             ewitab           = _mm_cvttpd_epi32(ewrt);
2463 #ifdef __XOP__
2464             eweps            = _mm_frcz_pd(ewrt);
2465 #else
2466             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2467 #endif
2468             twoeweps         = _mm_add_pd(eweps,eweps);
2469             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2470             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2471             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2472
2473             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
2474
2475             fscal            = felec;
2476
2477             fscal            = _mm_and_pd(fscal,cutoff_mask);
2478
2479             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2480
2481             /* Update vectorial force */
2482             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2483             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2484             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2485
2486             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2487             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2488             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2489
2490             }
2491
2492             /**************************
2493              * CALCULATE INTERACTIONS *
2494              **************************/
2495
2496             if (gmx_mm_any_lt(rsq23,rcutoff2))
2497             {
2498
2499             r23              = _mm_mul_pd(rsq23,rinv23);
2500
2501             /* EWALD ELECTROSTATICS */
2502
2503             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2504             ewrt             = _mm_mul_pd(r23,ewtabscale);
2505             ewitab           = _mm_cvttpd_epi32(ewrt);
2506 #ifdef __XOP__
2507             eweps            = _mm_frcz_pd(ewrt);
2508 #else
2509             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2510 #endif
2511             twoeweps         = _mm_add_pd(eweps,eweps);
2512             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2513             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2514             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
2515
2516             cutoff_mask      = _mm_cmplt_pd(rsq23,rcutoff2);
2517
2518             fscal            = felec;
2519
2520             fscal            = _mm_and_pd(fscal,cutoff_mask);
2521
2522             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2523
2524             /* Update vectorial force */
2525             fix2             = _mm_macc_pd(dx23,fscal,fix2);
2526             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
2527             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
2528
2529             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
2530             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
2531             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
2532
2533             }
2534
2535             /**************************
2536              * CALCULATE INTERACTIONS *
2537              **************************/
2538
2539             if (gmx_mm_any_lt(rsq31,rcutoff2))
2540             {
2541
2542             r31              = _mm_mul_pd(rsq31,rinv31);
2543
2544             /* EWALD ELECTROSTATICS */
2545
2546             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2547             ewrt             = _mm_mul_pd(r31,ewtabscale);
2548             ewitab           = _mm_cvttpd_epi32(ewrt);
2549 #ifdef __XOP__
2550             eweps            = _mm_frcz_pd(ewrt);
2551 #else
2552             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2553 #endif
2554             twoeweps         = _mm_add_pd(eweps,eweps);
2555             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2556             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2557             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2558
2559             cutoff_mask      = _mm_cmplt_pd(rsq31,rcutoff2);
2560
2561             fscal            = felec;
2562
2563             fscal            = _mm_and_pd(fscal,cutoff_mask);
2564
2565             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2566
2567             /* Update vectorial force */
2568             fix3             = _mm_macc_pd(dx31,fscal,fix3);
2569             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
2570             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
2571
2572             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
2573             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
2574             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
2575
2576             }
2577
2578             /**************************
2579              * CALCULATE INTERACTIONS *
2580              **************************/
2581
2582             if (gmx_mm_any_lt(rsq32,rcutoff2))
2583             {
2584
2585             r32              = _mm_mul_pd(rsq32,rinv32);
2586
2587             /* EWALD ELECTROSTATICS */
2588
2589             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2590             ewrt             = _mm_mul_pd(r32,ewtabscale);
2591             ewitab           = _mm_cvttpd_epi32(ewrt);
2592 #ifdef __XOP__
2593             eweps            = _mm_frcz_pd(ewrt);
2594 #else
2595             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2596 #endif
2597             twoeweps         = _mm_add_pd(eweps,eweps);
2598             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2599             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2600             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2601
2602             cutoff_mask      = _mm_cmplt_pd(rsq32,rcutoff2);
2603
2604             fscal            = felec;
2605
2606             fscal            = _mm_and_pd(fscal,cutoff_mask);
2607
2608             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2609
2610             /* Update vectorial force */
2611             fix3             = _mm_macc_pd(dx32,fscal,fix3);
2612             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
2613             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
2614
2615             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
2616             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
2617             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
2618
2619             }
2620
2621             /**************************
2622              * CALCULATE INTERACTIONS *
2623              **************************/
2624
2625             if (gmx_mm_any_lt(rsq33,rcutoff2))
2626             {
2627
2628             r33              = _mm_mul_pd(rsq33,rinv33);
2629
2630             /* EWALD ELECTROSTATICS */
2631
2632             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2633             ewrt             = _mm_mul_pd(r33,ewtabscale);
2634             ewitab           = _mm_cvttpd_epi32(ewrt);
2635 #ifdef __XOP__
2636             eweps            = _mm_frcz_pd(ewrt);
2637 #else
2638             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2639 #endif
2640             twoeweps         = _mm_add_pd(eweps,eweps);
2641             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2642             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2643             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2644
2645             cutoff_mask      = _mm_cmplt_pd(rsq33,rcutoff2);
2646
2647             fscal            = felec;
2648
2649             fscal            = _mm_and_pd(fscal,cutoff_mask);
2650
2651             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2652
2653             /* Update vectorial force */
2654             fix3             = _mm_macc_pd(dx33,fscal,fix3);
2655             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
2656             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
2657
2658             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
2659             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
2660             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
2661
2662             }
2663
2664             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2665
2666             /* Inner loop uses 414 flops */
2667         }
2668
2669         /* End of innermost loop */
2670
2671         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2672                                               f+i_coord_offset,fshift+i_shift_offset);
2673
2674         /* Increment number of inner iterations */
2675         inneriter                  += j_index_end - j_index_start;
2676
2677         /* Outer loop uses 24 flops */
2678     }
2679
2680     /* Increment number of outer iterations */
2681     outeriter        += nri;
2682
2683     /* Update outer/inner flops */
2684
2685     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*414);
2686 }