src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJ_GeomW4W4_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4W4_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW4W4_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     int              vdwjidx1A,vdwjidx1B;
  93     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  94     int              vdwjidx2A,vdwjidx2B;
  95     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  96     int              vdwjidx3A,vdwjidx3B;
  97     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  98     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 100     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 101     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 102     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 103     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 104     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 105     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 106     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 107     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 108     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 109     real             *charge;
 110     int              nvdwtype;
 111     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 112     int              *vdwtype;
 113     real             *vdwparam;
 114     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 115     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 116     __m128i          ewitab;
 117     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 118     real             *ewtab;
 119     __m128d          dummy_mask,cutoff_mask;
 120     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 121     __m128d          one     = _mm_set1_pd(1.0);
 122     __m128d          two     = _mm_set1_pd(2.0);
 123     x                = xx[0];
 124     f                = ff[0];
 125
 126     nri              = nlist->nri;
 127     iinr             = nlist->iinr;
 128     jindex           = nlist->jindex;
 129     jjnr             = nlist->jjnr;
 130     shiftidx         = nlist->shift;
 131     gid              = nlist->gid;
 132     shiftvec         = fr->shift_vec[0];
 133     fshift           = fr->fshift[0];
 134     facel            = _mm_set1_pd(fr->epsfac);
 135     charge           = mdatoms->chargeA;
 136     nvdwtype         = fr->ntype;
 137     vdwparam         = fr->nbfp;
 138     vdwtype          = mdatoms->typeA;
 139
 140     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 141     ewtab            = fr->ic->tabq_coul_FDV0;
 142     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 143     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 148     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 149     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 150     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 151
 152     jq1              = _mm_set1_pd(charge[inr+1]);
 153     jq2              = _mm_set1_pd(charge[inr+2]);
 154     jq3              = _mm_set1_pd(charge[inr+3]);
 155     vdwjidx0A        = 2*vdwtype[inr+0];
 156     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 157     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 158     qq11             = _mm_mul_pd(iq1,jq1);
 159     qq12             = _mm_mul_pd(iq1,jq2);
 160     qq13             = _mm_mul_pd(iq1,jq3);
 161     qq21             = _mm_mul_pd(iq2,jq1);
 162     qq22             = _mm_mul_pd(iq2,jq2);
 163     qq23             = _mm_mul_pd(iq2,jq3);
 164     qq31             = _mm_mul_pd(iq3,jq1);
 165     qq32             = _mm_mul_pd(iq3,jq2);
 166     qq33             = _mm_mul_pd(iq3,jq3);
 167
 168     /* Avoid stupid compiler warnings */
 169     jnrA = jnrB = 0;
 170     j_coord_offsetA = 0;
 171     j_coord_offsetB = 0;
 172
 173     outeriter        = 0;
 174     inneriter        = 0;
 175
 176     /* Start outer loop over neighborlists */
 177     for(iidx=0; iidx<nri; iidx++)
 178     {
 179         /* Load shift vector for this list */
 180         i_shift_offset   = DIM*shiftidx[iidx];
 181
 182         /* Load limits for loop over neighbors */
 183         j_index_start    = jindex[iidx];
 184         j_index_end      = jindex[iidx+1];
 185
 186         /* Get outer coordinate index */
 187         inr              = iinr[iidx];
 188         i_coord_offset   = DIM*inr;
 189
 190         /* Load i particle coords and add shift vector */
 191         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 192                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 193
 194         fix0             = _mm_setzero_pd();
 195         fiy0             = _mm_setzero_pd();
 196         fiz0             = _mm_setzero_pd();
 197         fix1             = _mm_setzero_pd();
 198         fiy1             = _mm_setzero_pd();
 199         fiz1             = _mm_setzero_pd();
 200         fix2             = _mm_setzero_pd();
 201         fiy2             = _mm_setzero_pd();
 202         fiz2             = _mm_setzero_pd();
 203         fix3             = _mm_setzero_pd();
 204         fiy3             = _mm_setzero_pd();
 205         fiz3             = _mm_setzero_pd();
 206
 207         /* Reset potential sums */
 208         velecsum         = _mm_setzero_pd();
 209         vvdwsum          = _mm_setzero_pd();
 210
 211         /* Start inner kernel loop */
 212         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 213         {
 214
 215             /* Get j neighbor index, and coordinate index */
 216             jnrA             = jjnr[jidx];
 217             jnrB             = jjnr[jidx+1];
 218             j_coord_offsetA  = DIM*jnrA;
 219             j_coord_offsetB  = DIM*jnrB;
 220
 221             /* load j atom coordinates */
 222             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 223                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 224                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm_sub_pd(ix0,jx0);
 228             dy00             = _mm_sub_pd(iy0,jy0);
 229             dz00             = _mm_sub_pd(iz0,jz0);
 230             dx11             = _mm_sub_pd(ix1,jx1);
 231             dy11             = _mm_sub_pd(iy1,jy1);
 232             dz11             = _mm_sub_pd(iz1,jz1);
 233             dx12             = _mm_sub_pd(ix1,jx2);
 234             dy12             = _mm_sub_pd(iy1,jy2);
 235             dz12             = _mm_sub_pd(iz1,jz2);
 236             dx13             = _mm_sub_pd(ix1,jx3);
 237             dy13             = _mm_sub_pd(iy1,jy3);
 238             dz13             = _mm_sub_pd(iz1,jz3);
 239             dx21             = _mm_sub_pd(ix2,jx1);
 240             dy21             = _mm_sub_pd(iy2,jy1);
 241             dz21             = _mm_sub_pd(iz2,jz1);
 242             dx22             = _mm_sub_pd(ix2,jx2);
 243             dy22             = _mm_sub_pd(iy2,jy2);
 244             dz22             = _mm_sub_pd(iz2,jz2);
 245             dx23             = _mm_sub_pd(ix2,jx3);
 246             dy23             = _mm_sub_pd(iy2,jy3);
 247             dz23             = _mm_sub_pd(iz2,jz3);
 248             dx31             = _mm_sub_pd(ix3,jx1);
 249             dy31             = _mm_sub_pd(iy3,jy1);
 250             dz31             = _mm_sub_pd(iz3,jz1);
 251             dx32             = _mm_sub_pd(ix3,jx2);
 252             dy32             = _mm_sub_pd(iy3,jy2);
 253             dz32             = _mm_sub_pd(iz3,jz2);
 254             dx33             = _mm_sub_pd(ix3,jx3);
 255             dy33             = _mm_sub_pd(iy3,jy3);
 256             dz33             = _mm_sub_pd(iz3,jz3);
 257
 258             /* Calculate squared distance and things based on it */
 259             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 260             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 261             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 262             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 263             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 264             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 265             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 266             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 267             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 268             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 269
 270             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 271             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 272             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 273             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 274             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 275             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 276             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 277             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 278             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 279
 280             rinvsq00         = gmx_mm_inv_pd(rsq00);
 281             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 282             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 283             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 284             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 285             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 286             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 287             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 288             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 289             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 290
 291             fjx0             = _mm_setzero_pd();
 292             fjy0             = _mm_setzero_pd();
 293             fjz0             = _mm_setzero_pd();
 294             fjx1             = _mm_setzero_pd();
 295             fjy1             = _mm_setzero_pd();
 296             fjz1             = _mm_setzero_pd();
 297             fjx2             = _mm_setzero_pd();
 298             fjy2             = _mm_setzero_pd();
 299             fjz2             = _mm_setzero_pd();
 300             fjx3             = _mm_setzero_pd();
 301             fjy3             = _mm_setzero_pd();
 302             fjz3             = _mm_setzero_pd();
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             /* LENNARD-JONES DISPERSION/REPULSION */
 309
 310             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 311             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 312             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 313             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 314             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 318
 319             fscal            = fvdw;
 320
 321             /* Update vectorial force */
 322             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 323             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 324             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 325
 326             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 327             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 328             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             r11              = _mm_mul_pd(rsq11,rinv11);
 335
 336             /* EWALD ELECTROSTATICS */
 337
 338             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 339             ewrt             = _mm_mul_pd(r11,ewtabscale);
 340             ewitab           = _mm_cvttpd_epi32(ewrt);
 341 #ifdef __XOP__
 342             eweps            = _mm_frcz_pd(ewrt);
 343 #else
 344             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 345 #endif
 346             twoeweps         = _mm_add_pd(eweps,eweps);
 347             ewitab           = _mm_slli_epi32(ewitab,2);
 348             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 349             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 350             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 351             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 352             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 353             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 354             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 355             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 356             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 357             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velecsum         = _mm_add_pd(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             /* Update vectorial force */
 365             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 366             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 367             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 368
 369             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 370             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 371             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 372
 373             /**************************
 374              * CALCULATE INTERACTIONS *
 375              **************************/
 376
 377             r12              = _mm_mul_pd(rsq12,rinv12);
 378
 379             /* EWALD ELECTROSTATICS */
 380
 381             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 382             ewrt             = _mm_mul_pd(r12,ewtabscale);
 383             ewitab           = _mm_cvttpd_epi32(ewrt);
 384 #ifdef __XOP__
 385             eweps            = _mm_frcz_pd(ewrt);
 386 #else
 387             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 388 #endif
 389             twoeweps         = _mm_add_pd(eweps,eweps);
 390             ewitab           = _mm_slli_epi32(ewitab,2);
 391             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 392             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 393             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 394             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 395             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 396             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 397             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 398             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 399             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 400             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 401
 402             /* Update potential sum for this i atom from the interaction with this j atom. */
 403             velecsum         = _mm_add_pd(velecsum,velec);
 404
 405             fscal            = felec;
 406
 407             /* Update vectorial force */
 408             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 409             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 410             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 411
 412             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 413             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 414             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 415
 416             /**************************
 417              * CALCULATE INTERACTIONS *
 418              **************************/
 419
 420             r13              = _mm_mul_pd(rsq13,rinv13);
 421
 422             /* EWALD ELECTROSTATICS */
 423
 424             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 425             ewrt             = _mm_mul_pd(r13,ewtabscale);
 426             ewitab           = _mm_cvttpd_epi32(ewrt);
 427 #ifdef __XOP__
 428             eweps            = _mm_frcz_pd(ewrt);
 429 #else
 430             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 431 #endif
 432             twoeweps         = _mm_add_pd(eweps,eweps);
 433             ewitab           = _mm_slli_epi32(ewitab,2);
 434             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 435             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 436             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 437             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 438             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 439             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 440             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 441             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 442             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
 443             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 444
 445             /* Update potential sum for this i atom from the interaction with this j atom. */
 446             velecsum         = _mm_add_pd(velecsum,velec);
 447
 448             fscal            = felec;
 449
 450             /* Update vectorial force */
 451             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 452             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 453             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 454
 455             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 456             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 457             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             r21              = _mm_mul_pd(rsq21,rinv21);
 464
 465             /* EWALD ELECTROSTATICS */
 466
 467             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 468             ewrt             = _mm_mul_pd(r21,ewtabscale);
 469             ewitab           = _mm_cvttpd_epi32(ewrt);
 470 #ifdef __XOP__
 471             eweps            = _mm_frcz_pd(ewrt);
 472 #else
 473             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 474 #endif
 475             twoeweps         = _mm_add_pd(eweps,eweps);
 476             ewitab           = _mm_slli_epi32(ewitab,2);
 477             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 478             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 479             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 480             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 481             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 482             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 483             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 484             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 485             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 486             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velecsum         = _mm_add_pd(velecsum,velec);
 490
 491             fscal            = felec;
 492
 493             /* Update vectorial force */
 494             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 495             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 496             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 497
 498             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 499             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 500             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             r22              = _mm_mul_pd(rsq22,rinv22);
 507
 508             /* EWALD ELECTROSTATICS */
 509
 510             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 511             ewrt             = _mm_mul_pd(r22,ewtabscale);
 512             ewitab           = _mm_cvttpd_epi32(ewrt);
 513 #ifdef __XOP__
 514             eweps            = _mm_frcz_pd(ewrt);
 515 #else
 516             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 517 #endif
 518             twoeweps         = _mm_add_pd(eweps,eweps);
 519             ewitab           = _mm_slli_epi32(ewitab,2);
 520             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 521             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 522             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 523             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 524             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 525             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 526             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 527             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 528             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 529             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 530
 531             /* Update potential sum for this i atom from the interaction with this j atom. */
 532             velecsum         = _mm_add_pd(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             /* Update vectorial force */
 537             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 538             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 539             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 540
 541             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 542             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 543             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             r23              = _mm_mul_pd(rsq23,rinv23);
 550
 551             /* EWALD ELECTROSTATICS */
 552
 553             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 554             ewrt             = _mm_mul_pd(r23,ewtabscale);
 555             ewitab           = _mm_cvttpd_epi32(ewrt);
 556 #ifdef __XOP__
 557             eweps            = _mm_frcz_pd(ewrt);
 558 #else
 559             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 560 #endif
 561             twoeweps         = _mm_add_pd(eweps,eweps);
 562             ewitab           = _mm_slli_epi32(ewitab,2);
 563             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 564             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 565             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 566             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 567             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 568             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 569             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 570             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 571             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
 572             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velecsum         = _mm_add_pd(velecsum,velec);
 576
 577             fscal            = felec;
 578
 579             /* Update vectorial force */
 580             fix2             = _mm_macc_pd(dx23,fscal,fix2);
 581             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
 582             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
 583
 584             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
 585             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
 586             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
 587
 588             /**************************
 589              * CALCULATE INTERACTIONS *
 590              **************************/
 591
 592             r31              = _mm_mul_pd(rsq31,rinv31);
 593
 594             /* EWALD ELECTROSTATICS */
 595
 596             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 597             ewrt             = _mm_mul_pd(r31,ewtabscale);
 598             ewitab           = _mm_cvttpd_epi32(ewrt);
 599 #ifdef __XOP__
 600             eweps            = _mm_frcz_pd(ewrt);
 601 #else
 602             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 603 #endif
 604             twoeweps         = _mm_add_pd(eweps,eweps);
 605             ewitab           = _mm_slli_epi32(ewitab,2);
 606             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 607             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 608             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 609             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 610             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 611             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 612             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 613             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 614             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
 615             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
 616
 617             /* Update potential sum for this i atom from the interaction with this j atom. */
 618             velecsum         = _mm_add_pd(velecsum,velec);
 619
 620             fscal            = felec;
 621
 622             /* Update vectorial force */
 623             fix3             = _mm_macc_pd(dx31,fscal,fix3);
 624             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
 625             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
 626
 627             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
 628             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
 629             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
 630
 631             /**************************
 632              * CALCULATE INTERACTIONS *
 633              **************************/
 634
 635             r32              = _mm_mul_pd(rsq32,rinv32);
 636
 637             /* EWALD ELECTROSTATICS */
 638
 639             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 640             ewrt             = _mm_mul_pd(r32,ewtabscale);
 641             ewitab           = _mm_cvttpd_epi32(ewrt);
 642 #ifdef __XOP__
 643             eweps            = _mm_frcz_pd(ewrt);
 644 #else
 645             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 646 #endif
 647             twoeweps         = _mm_add_pd(eweps,eweps);
 648             ewitab           = _mm_slli_epi32(ewitab,2);
 649             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 650             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 651             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 652             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 653             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 654             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 655             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 656             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 657             velec            = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
 658             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
 659
 660             /* Update potential sum for this i atom from the interaction with this j atom. */
 661             velecsum         = _mm_add_pd(velecsum,velec);
 662
 663             fscal            = felec;
 664
 665             /* Update vectorial force */
 666             fix3             = _mm_macc_pd(dx32,fscal,fix3);
 667             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
 668             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
 669
 670             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
 671             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
 672             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
 673
 674             /**************************
 675              * CALCULATE INTERACTIONS *
 676              **************************/
 677
 678             r33              = _mm_mul_pd(rsq33,rinv33);
 679
 680             /* EWALD ELECTROSTATICS */
 681
 682             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 683             ewrt             = _mm_mul_pd(r33,ewtabscale);
 684             ewitab           = _mm_cvttpd_epi32(ewrt);
 685 #ifdef __XOP__
 686             eweps            = _mm_frcz_pd(ewrt);
 687 #else
 688             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 689 #endif
 690             twoeweps         = _mm_add_pd(eweps,eweps);
 691             ewitab           = _mm_slli_epi32(ewitab,2);
 692             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 693             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 694             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 695             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 696             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 697             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 698             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 699             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 700             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
 701             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
 702
 703             /* Update potential sum for this i atom from the interaction with this j atom. */
 704             velecsum         = _mm_add_pd(velecsum,velec);
 705
 706             fscal            = felec;
 707
 708             /* Update vectorial force */
 709             fix3             = _mm_macc_pd(dx33,fscal,fix3);
 710             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
 711             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
 712
 713             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
 714             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
 715             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
 716
 717             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);
 718
 719             /* Inner loop uses 434 flops */
 720         }
 721
 722         if(jidx<j_index_end)
 723         {
 724
 725             jnrA             = jjnr[jidx];
 726             j_coord_offsetA  = DIM*jnrA;
 727
 728             /* load j atom coordinates */
 729             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 730                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 731                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 732
 733             /* Calculate displacement vector */
 734             dx00             = _mm_sub_pd(ix0,jx0);
 735             dy00             = _mm_sub_pd(iy0,jy0);
 736             dz00             = _mm_sub_pd(iz0,jz0);
 737             dx11             = _mm_sub_pd(ix1,jx1);
 738             dy11             = _mm_sub_pd(iy1,jy1);
 739             dz11             = _mm_sub_pd(iz1,jz1);
 740             dx12             = _mm_sub_pd(ix1,jx2);
 741             dy12             = _mm_sub_pd(iy1,jy2);
 742             dz12             = _mm_sub_pd(iz1,jz2);
 743             dx13             = _mm_sub_pd(ix1,jx3);
 744             dy13             = _mm_sub_pd(iy1,jy3);
 745             dz13             = _mm_sub_pd(iz1,jz3);
 746             dx21             = _mm_sub_pd(ix2,jx1);
 747             dy21             = _mm_sub_pd(iy2,jy1);
 748             dz21             = _mm_sub_pd(iz2,jz1);
 749             dx22             = _mm_sub_pd(ix2,jx2);
 750             dy22             = _mm_sub_pd(iy2,jy2);
 751             dz22             = _mm_sub_pd(iz2,jz2);
 752             dx23             = _mm_sub_pd(ix2,jx3);
 753             dy23             = _mm_sub_pd(iy2,jy3);
 754             dz23             = _mm_sub_pd(iz2,jz3);
 755             dx31             = _mm_sub_pd(ix3,jx1);
 756             dy31             = _mm_sub_pd(iy3,jy1);
 757             dz31             = _mm_sub_pd(iz3,jz1);
 758             dx32             = _mm_sub_pd(ix3,jx2);
 759             dy32             = _mm_sub_pd(iy3,jy2);
 760             dz32             = _mm_sub_pd(iz3,jz2);
 761             dx33             = _mm_sub_pd(ix3,jx3);
 762             dy33             = _mm_sub_pd(iy3,jy3);
 763             dz33             = _mm_sub_pd(iz3,jz3);
 764
 765             /* Calculate squared distance and things based on it */
 766             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 767             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 768             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 769             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 770             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 771             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 772             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 773             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 774             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 775             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 776
 777             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 778             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 779             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 780             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 781             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 782             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 783             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 784             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 785             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 786
 787             rinvsq00         = gmx_mm_inv_pd(rsq00);
 788             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 789             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 790             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 791             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 792             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 793             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 794             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 795             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 796             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 797
 798             fjx0             = _mm_setzero_pd();
 799             fjy0             = _mm_setzero_pd();
 800             fjz0             = _mm_setzero_pd();
 801             fjx1             = _mm_setzero_pd();
 802             fjy1             = _mm_setzero_pd();
 803             fjz1             = _mm_setzero_pd();
 804             fjx2             = _mm_setzero_pd();
 805             fjy2             = _mm_setzero_pd();
 806             fjz2             = _mm_setzero_pd();
 807             fjx3             = _mm_setzero_pd();
 808             fjy3             = _mm_setzero_pd();
 809             fjz3             = _mm_setzero_pd();
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             /* LENNARD-JONES DISPERSION/REPULSION */
 816
 817             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 818             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 819             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 820             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 821             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 822
 823             /* Update potential sum for this i atom from the interaction with this j atom. */
 824             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 825             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 826
 827             fscal            = fvdw;
 828
 829             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 830
 831             /* Update vectorial force */
 832             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 833             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 834             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 835
 836             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 837             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 838             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 839
 840             /**************************
 841              * CALCULATE INTERACTIONS *
 842              **************************/
 843
 844             r11              = _mm_mul_pd(rsq11,rinv11);
 845
 846             /* EWALD ELECTROSTATICS */
 847
 848             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 849             ewrt             = _mm_mul_pd(r11,ewtabscale);
 850             ewitab           = _mm_cvttpd_epi32(ewrt);
 851 #ifdef __XOP__
 852             eweps            = _mm_frcz_pd(ewrt);
 853 #else
 854             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 855 #endif
 856             twoeweps         = _mm_add_pd(eweps,eweps);
 857             ewitab           = _mm_slli_epi32(ewitab,2);
 858             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 859             ewtabD           = _mm_setzero_pd();
 860             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 861             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 862             ewtabFn          = _mm_setzero_pd();
 863             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 864             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 865             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 866             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 867             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 868
 869             /* Update potential sum for this i atom from the interaction with this j atom. */
 870             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 871             velecsum         = _mm_add_pd(velecsum,velec);
 872
 873             fscal            = felec;
 874
 875             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 876
 877             /* Update vectorial force */
 878             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 879             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 880             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 881
 882             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 883             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 884             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 885
 886             /**************************
 887              * CALCULATE INTERACTIONS *
 888              **************************/
 889
 890             r12              = _mm_mul_pd(rsq12,rinv12);
 891
 892             /* EWALD ELECTROSTATICS */
 893
 894             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 895             ewrt             = _mm_mul_pd(r12,ewtabscale);
 896             ewitab           = _mm_cvttpd_epi32(ewrt);
 897 #ifdef __XOP__
 898             eweps            = _mm_frcz_pd(ewrt);
 899 #else
 900             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 901 #endif
 902             twoeweps         = _mm_add_pd(eweps,eweps);
 903             ewitab           = _mm_slli_epi32(ewitab,2);
 904             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 905             ewtabD           = _mm_setzero_pd();
 906             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 907             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 908             ewtabFn          = _mm_setzero_pd();
 909             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 910             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 911             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 912             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 913             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 914
 915             /* Update potential sum for this i atom from the interaction with this j atom. */
 916             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 917             velecsum         = _mm_add_pd(velecsum,velec);
 918
 919             fscal            = felec;
 920
 921             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 922
 923             /* Update vectorial force */
 924             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 925             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 926             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 927
 928             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 929             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 930             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 931
 932             /**************************
 933              * CALCULATE INTERACTIONS *
 934              **************************/
 935
 936             r13              = _mm_mul_pd(rsq13,rinv13);
 937
 938             /* EWALD ELECTROSTATICS */
 939
 940             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 941             ewrt             = _mm_mul_pd(r13,ewtabscale);
 942             ewitab           = _mm_cvttpd_epi32(ewrt);
 943 #ifdef __XOP__
 944             eweps            = _mm_frcz_pd(ewrt);
 945 #else
 946             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 947 #endif
 948             twoeweps         = _mm_add_pd(eweps,eweps);
 949             ewitab           = _mm_slli_epi32(ewitab,2);
 950             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 951             ewtabD           = _mm_setzero_pd();
 952             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 953             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 954             ewtabFn          = _mm_setzero_pd();
 955             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 956             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 957             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 958             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
 959             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 960
 961             /* Update potential sum for this i atom from the interaction with this j atom. */
 962             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 963             velecsum         = _mm_add_pd(velecsum,velec);
 964
 965             fscal            = felec;
 966
 967             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 968
 969             /* Update vectorial force */
 970             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 971             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 972             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 973
 974             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 975             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 976             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             r21              = _mm_mul_pd(rsq21,rinv21);
 983
 984             /* EWALD ELECTROSTATICS */
 985
 986             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 987             ewrt             = _mm_mul_pd(r21,ewtabscale);
 988             ewitab           = _mm_cvttpd_epi32(ewrt);
 989 #ifdef __XOP__
 990             eweps            = _mm_frcz_pd(ewrt);
 991 #else
 992             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 993 #endif
 994             twoeweps         = _mm_add_pd(eweps,eweps);
 995             ewitab           = _mm_slli_epi32(ewitab,2);
 996             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 997             ewtabD           = _mm_setzero_pd();
 998             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 999             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1000             ewtabFn          = _mm_setzero_pd();
1001             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1002             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1003             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1004             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1005             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1006
1007             /* Update potential sum for this i atom from the interaction with this j atom. */
1008             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1009             velecsum         = _mm_add_pd(velecsum,velec);
1010
1011             fscal            = felec;
1012
1013             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1014
1015             /* Update vectorial force */
1016             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1017             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1018             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1019
1020             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1021             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1022             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1023
1024             /**************************
1025              * CALCULATE INTERACTIONS *
1026              **************************/
1027
1028             r22              = _mm_mul_pd(rsq22,rinv22);
1029
1030             /* EWALD ELECTROSTATICS */
1031
1032             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1033             ewrt             = _mm_mul_pd(r22,ewtabscale);
1034             ewitab           = _mm_cvttpd_epi32(ewrt);
1035 #ifdef __XOP__
1036             eweps            = _mm_frcz_pd(ewrt);
1037 #else
1038             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1039 #endif
1040             twoeweps         = _mm_add_pd(eweps,eweps);
1041             ewitab           = _mm_slli_epi32(ewitab,2);
1042             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1043             ewtabD           = _mm_setzero_pd();
1044             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1045             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1046             ewtabFn          = _mm_setzero_pd();
1047             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1048             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1049             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1050             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1051             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1052
1053             /* Update potential sum for this i atom from the interaction with this j atom. */
1054             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1055             velecsum         = _mm_add_pd(velecsum,velec);
1056
1057             fscal            = felec;
1058
1059             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1060
1061             /* Update vectorial force */
1062             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1063             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1064             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1065
1066             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1067             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1068             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             r23              = _mm_mul_pd(rsq23,rinv23);
1075
1076             /* EWALD ELECTROSTATICS */
1077
1078             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1079             ewrt             = _mm_mul_pd(r23,ewtabscale);
1080             ewitab           = _mm_cvttpd_epi32(ewrt);
1081 #ifdef __XOP__
1082             eweps            = _mm_frcz_pd(ewrt);
1083 #else
1084             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1085 #endif
1086             twoeweps         = _mm_add_pd(eweps,eweps);
1087             ewitab           = _mm_slli_epi32(ewitab,2);
1088             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1089             ewtabD           = _mm_setzero_pd();
1090             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1091             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1092             ewtabFn          = _mm_setzero_pd();
1093             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1094             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1095             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1096             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
1097             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1098
1099             /* Update potential sum for this i atom from the interaction with this j atom. */
1100             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1101             velecsum         = _mm_add_pd(velecsum,velec);
1102
1103             fscal            = felec;
1104
1105             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1106
1107             /* Update vectorial force */
1108             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1109             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1110             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1111
1112             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1113             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1114             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1115
1116             /**************************
1117              * CALCULATE INTERACTIONS *
1118              **************************/
1119
1120             r31              = _mm_mul_pd(rsq31,rinv31);
1121
1122             /* EWALD ELECTROSTATICS */
1123
1124             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1125             ewrt             = _mm_mul_pd(r31,ewtabscale);
1126             ewitab           = _mm_cvttpd_epi32(ewrt);
1127 #ifdef __XOP__
1128             eweps            = _mm_frcz_pd(ewrt);
1129 #else
1130             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1131 #endif
1132             twoeweps         = _mm_add_pd(eweps,eweps);
1133             ewitab           = _mm_slli_epi32(ewitab,2);
1134             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1135             ewtabD           = _mm_setzero_pd();
1136             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1137             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1138             ewtabFn          = _mm_setzero_pd();
1139             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1140             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1141             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1142             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
1143             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1144
1145             /* Update potential sum for this i atom from the interaction with this j atom. */
1146             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1147             velecsum         = _mm_add_pd(velecsum,velec);
1148
1149             fscal            = felec;
1150
1151             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1152
1153             /* Update vectorial force */
1154             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1155             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1156             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1157
1158             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1159             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1160             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1161
1162             /**************************
1163              * CALCULATE INTERACTIONS *
1164              **************************/
1165
1166             r32              = _mm_mul_pd(rsq32,rinv32);
1167
1168             /* EWALD ELECTROSTATICS */
1169
1170             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1171             ewrt             = _mm_mul_pd(r32,ewtabscale);
1172             ewitab           = _mm_cvttpd_epi32(ewrt);
1173 #ifdef __XOP__
1174             eweps            = _mm_frcz_pd(ewrt);
1175 #else
1176             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1177 #endif
1178             twoeweps         = _mm_add_pd(eweps,eweps);
1179             ewitab           = _mm_slli_epi32(ewitab,2);
1180             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1181             ewtabD           = _mm_setzero_pd();
1182             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1183             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1184             ewtabFn          = _mm_setzero_pd();
1185             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1186             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1187             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1188             velec            = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
1189             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1190
1191             /* Update potential sum for this i atom from the interaction with this j atom. */
1192             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1193             velecsum         = _mm_add_pd(velecsum,velec);
1194
1195             fscal            = felec;
1196
1197             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1198
1199             /* Update vectorial force */
1200             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1201             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1202             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1203
1204             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1205             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1206             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1207
1208             /**************************
1209              * CALCULATE INTERACTIONS *
1210              **************************/
1211
1212             r33              = _mm_mul_pd(rsq33,rinv33);
1213
1214             /* EWALD ELECTROSTATICS */
1215
1216             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1217             ewrt             = _mm_mul_pd(r33,ewtabscale);
1218             ewitab           = _mm_cvttpd_epi32(ewrt);
1219 #ifdef __XOP__
1220             eweps            = _mm_frcz_pd(ewrt);
1221 #else
1222             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1223 #endif
1224             twoeweps         = _mm_add_pd(eweps,eweps);
1225             ewitab           = _mm_slli_epi32(ewitab,2);
1226             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1227             ewtabD           = _mm_setzero_pd();
1228             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1229             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1230             ewtabFn          = _mm_setzero_pd();
1231             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1232             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1233             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1234             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
1235             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1236
1237             /* Update potential sum for this i atom from the interaction with this j atom. */
1238             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1239             velecsum         = _mm_add_pd(velecsum,velec);
1240
1241             fscal            = felec;
1242
1243             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1244
1245             /* Update vectorial force */
1246             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1247             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1248             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1249
1250             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1251             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1252             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1253
1254             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1255
1256             /* Inner loop uses 434 flops */
1257         }
1258
1259         /* End of innermost loop */
1260
1261         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1262                                               f+i_coord_offset,fshift+i_shift_offset);
1263
1264         ggid                        = gid[iidx];
1265         /* Update potential energies */
1266         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1267         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1268
1269         /* Increment number of inner iterations */
1270         inneriter                  += j_index_end - j_index_start;
1271
1272         /* Outer loop uses 26 flops */
1273     }
1274
1275     /* Increment number of outer iterations */
1276     outeriter        += nri;
1277
1278     /* Update outer/inner flops */
1279
1280     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*434);
1281 }
1282 /*
1283  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4W4_F_avx_128_fma_double
1284  * Electrostatics interaction: Ewald
1285  * VdW interaction:            LennardJones
1286  * Geometry:                   Water4-Water4
1287  * Calculate force/pot:        Force
1288  */
1289 void
1290 nb_kernel_ElecEw_VdwLJ_GeomW4W4_F_avx_128_fma_double
1291                     (t_nblist                    * gmx_restrict       nlist,
1292                      rvec                        * gmx_restrict          xx,
1293                      rvec                        * gmx_restrict          ff,
1294                      t_forcerec                  * gmx_restrict          fr,
1295                      t_mdatoms                   * gmx_restrict     mdatoms,
1296                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1297                      t_nrnb                      * gmx_restrict        nrnb)
1298 {
1299     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1300      * just 0 for non-waters.
1301      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1302      * jnr indices corresponding to data put in the four positions in the SIMD register.
1303      */
1304     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1305     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1306     int              jnrA,jnrB;
1307     int              j_coord_offsetA,j_coord_offsetB;
1308     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1309     real             rcutoff_scalar;
1310     real             *shiftvec,*fshift,*x,*f;
1311     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1312     int              vdwioffset0;
1313     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1314     int              vdwioffset1;
1315     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1316     int              vdwioffset2;
1317     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1318     int              vdwioffset3;
1319     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1320     int              vdwjidx0A,vdwjidx0B;
1321     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1322     int              vdwjidx1A,vdwjidx1B;
1323     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1324     int              vdwjidx2A,vdwjidx2B;
1325     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1326     int              vdwjidx3A,vdwjidx3B;
1327     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1328     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1329     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1330     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1331     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1332     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1333     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1334     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1335     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1336     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1337     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1338     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1339     real             *charge;
1340     int              nvdwtype;
1341     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1342     int              *vdwtype;
1343     real             *vdwparam;
1344     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1345     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1346     __m128i          ewitab;
1347     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1348     real             *ewtab;
1349     __m128d          dummy_mask,cutoff_mask;
1350     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1351     __m128d          one     = _mm_set1_pd(1.0);
1352     __m128d          two     = _mm_set1_pd(2.0);
1353     x                = xx[0];
1354     f                = ff[0];
1355
1356     nri              = nlist->nri;
1357     iinr             = nlist->iinr;
1358     jindex           = nlist->jindex;
1359     jjnr             = nlist->jjnr;
1360     shiftidx         = nlist->shift;
1361     gid              = nlist->gid;
1362     shiftvec         = fr->shift_vec[0];
1363     fshift           = fr->fshift[0];
1364     facel            = _mm_set1_pd(fr->epsfac);
1365     charge           = mdatoms->chargeA;
1366     nvdwtype         = fr->ntype;
1367     vdwparam         = fr->nbfp;
1368     vdwtype          = mdatoms->typeA;
1369
1370     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1371     ewtab            = fr->ic->tabq_coul_F;
1372     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1373     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1374
1375     /* Setup water-specific parameters */
1376     inr              = nlist->iinr[0];
1377     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1378     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1379     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1380     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1381
1382     jq1              = _mm_set1_pd(charge[inr+1]);
1383     jq2              = _mm_set1_pd(charge[inr+2]);
1384     jq3              = _mm_set1_pd(charge[inr+3]);
1385     vdwjidx0A        = 2*vdwtype[inr+0];
1386     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1387     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1388     qq11             = _mm_mul_pd(iq1,jq1);
1389     qq12             = _mm_mul_pd(iq1,jq2);
1390     qq13             = _mm_mul_pd(iq1,jq3);
1391     qq21             = _mm_mul_pd(iq2,jq1);
1392     qq22             = _mm_mul_pd(iq2,jq2);
1393     qq23             = _mm_mul_pd(iq2,jq3);
1394     qq31             = _mm_mul_pd(iq3,jq1);
1395     qq32             = _mm_mul_pd(iq3,jq2);
1396     qq33             = _mm_mul_pd(iq3,jq3);
1397
1398     /* Avoid stupid compiler warnings */
1399     jnrA = jnrB = 0;
1400     j_coord_offsetA = 0;
1401     j_coord_offsetB = 0;
1402
1403     outeriter        = 0;
1404     inneriter        = 0;
1405
1406     /* Start outer loop over neighborlists */
1407     for(iidx=0; iidx<nri; iidx++)
1408     {
1409         /* Load shift vector for this list */
1410         i_shift_offset   = DIM*shiftidx[iidx];
1411
1412         /* Load limits for loop over neighbors */
1413         j_index_start    = jindex[iidx];
1414         j_index_end      = jindex[iidx+1];
1415
1416         /* Get outer coordinate index */
1417         inr              = iinr[iidx];
1418         i_coord_offset   = DIM*inr;
1419
1420         /* Load i particle coords and add shift vector */
1421         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1422                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1423
1424         fix0             = _mm_setzero_pd();
1425         fiy0             = _mm_setzero_pd();
1426         fiz0             = _mm_setzero_pd();
1427         fix1             = _mm_setzero_pd();
1428         fiy1             = _mm_setzero_pd();
1429         fiz1             = _mm_setzero_pd();
1430         fix2             = _mm_setzero_pd();
1431         fiy2             = _mm_setzero_pd();
1432         fiz2             = _mm_setzero_pd();
1433         fix3             = _mm_setzero_pd();
1434         fiy3             = _mm_setzero_pd();
1435         fiz3             = _mm_setzero_pd();
1436
1437         /* Start inner kernel loop */
1438         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1439         {
1440
1441             /* Get j neighbor index, and coordinate index */
1442             jnrA             = jjnr[jidx];
1443             jnrB             = jjnr[jidx+1];
1444             j_coord_offsetA  = DIM*jnrA;
1445             j_coord_offsetB  = DIM*jnrB;
1446
1447             /* load j atom coordinates */
1448             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1449                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1450                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1451
1452             /* Calculate displacement vector */
1453             dx00             = _mm_sub_pd(ix0,jx0);
1454             dy00             = _mm_sub_pd(iy0,jy0);
1455             dz00             = _mm_sub_pd(iz0,jz0);
1456             dx11             = _mm_sub_pd(ix1,jx1);
1457             dy11             = _mm_sub_pd(iy1,jy1);
1458             dz11             = _mm_sub_pd(iz1,jz1);
1459             dx12             = _mm_sub_pd(ix1,jx2);
1460             dy12             = _mm_sub_pd(iy1,jy2);
1461             dz12             = _mm_sub_pd(iz1,jz2);
1462             dx13             = _mm_sub_pd(ix1,jx3);
1463             dy13             = _mm_sub_pd(iy1,jy3);
1464             dz13             = _mm_sub_pd(iz1,jz3);
1465             dx21             = _mm_sub_pd(ix2,jx1);
1466             dy21             = _mm_sub_pd(iy2,jy1);
1467             dz21             = _mm_sub_pd(iz2,jz1);
1468             dx22             = _mm_sub_pd(ix2,jx2);
1469             dy22             = _mm_sub_pd(iy2,jy2);
1470             dz22             = _mm_sub_pd(iz2,jz2);
1471             dx23             = _mm_sub_pd(ix2,jx3);
1472             dy23             = _mm_sub_pd(iy2,jy3);
1473             dz23             = _mm_sub_pd(iz2,jz3);
1474             dx31             = _mm_sub_pd(ix3,jx1);
1475             dy31             = _mm_sub_pd(iy3,jy1);
1476             dz31             = _mm_sub_pd(iz3,jz1);
1477             dx32             = _mm_sub_pd(ix3,jx2);
1478             dy32             = _mm_sub_pd(iy3,jy2);
1479             dz32             = _mm_sub_pd(iz3,jz2);
1480             dx33             = _mm_sub_pd(ix3,jx3);
1481             dy33             = _mm_sub_pd(iy3,jy3);
1482             dz33             = _mm_sub_pd(iz3,jz3);
1483
1484             /* Calculate squared distance and things based on it */
1485             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1486             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1487             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1488             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1489             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1490             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1491             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1492             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1493             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1494             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1495
1496             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1497             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1498             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1499             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1500             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1501             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1502             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1503             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1504             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1505
1506             rinvsq00         = gmx_mm_inv_pd(rsq00);
1507             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1508             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1509             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1510             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1511             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1512             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1513             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1514             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1515             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1516
1517             fjx0             = _mm_setzero_pd();
1518             fjy0             = _mm_setzero_pd();
1519             fjz0             = _mm_setzero_pd();
1520             fjx1             = _mm_setzero_pd();
1521             fjy1             = _mm_setzero_pd();
1522             fjz1             = _mm_setzero_pd();
1523             fjx2             = _mm_setzero_pd();
1524             fjy2             = _mm_setzero_pd();
1525             fjz2             = _mm_setzero_pd();
1526             fjx3             = _mm_setzero_pd();
1527             fjy3             = _mm_setzero_pd();
1528             fjz3             = _mm_setzero_pd();
1529
1530             /**************************
1531              * CALCULATE INTERACTIONS *
1532              **************************/
1533
1534             /* LENNARD-JONES DISPERSION/REPULSION */
1535
1536             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1537             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1538
1539             fscal            = fvdw;
1540
1541             /* Update vectorial force */
1542             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1543             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1544             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1545
1546             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1547             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1548             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1549
1550             /**************************
1551              * CALCULATE INTERACTIONS *
1552              **************************/
1553
1554             r11              = _mm_mul_pd(rsq11,rinv11);
1555
1556             /* EWALD ELECTROSTATICS */
1557
1558             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1559             ewrt             = _mm_mul_pd(r11,ewtabscale);
1560             ewitab           = _mm_cvttpd_epi32(ewrt);
1561 #ifdef __XOP__
1562             eweps            = _mm_frcz_pd(ewrt);
1563 #else
1564             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1565 #endif
1566             twoeweps         = _mm_add_pd(eweps,eweps);
1567             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1568                                          &ewtabF,&ewtabFn);
1569             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1570             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1571
1572             fscal            = felec;
1573
1574             /* Update vectorial force */
1575             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1576             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1577             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1578
1579             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1580             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1581             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1582
1583             /**************************
1584              * CALCULATE INTERACTIONS *
1585              **************************/
1586
1587             r12              = _mm_mul_pd(rsq12,rinv12);
1588
1589             /* EWALD ELECTROSTATICS */
1590
1591             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1592             ewrt             = _mm_mul_pd(r12,ewtabscale);
1593             ewitab           = _mm_cvttpd_epi32(ewrt);
1594 #ifdef __XOP__
1595             eweps            = _mm_frcz_pd(ewrt);
1596 #else
1597             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1598 #endif
1599             twoeweps         = _mm_add_pd(eweps,eweps);
1600             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1601                                          &ewtabF,&ewtabFn);
1602             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1603             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1604
1605             fscal            = felec;
1606
1607             /* Update vectorial force */
1608             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1609             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1610             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1611
1612             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1613             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1614             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1615
1616             /**************************
1617              * CALCULATE INTERACTIONS *
1618              **************************/
1619
1620             r13              = _mm_mul_pd(rsq13,rinv13);
1621
1622             /* EWALD ELECTROSTATICS */
1623
1624             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1625             ewrt             = _mm_mul_pd(r13,ewtabscale);
1626             ewitab           = _mm_cvttpd_epi32(ewrt);
1627 #ifdef __XOP__
1628             eweps            = _mm_frcz_pd(ewrt);
1629 #else
1630             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1631 #endif
1632             twoeweps         = _mm_add_pd(eweps,eweps);
1633             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1634                                          &ewtabF,&ewtabFn);
1635             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1636             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1637
1638             fscal            = felec;
1639
1640             /* Update vectorial force */
1641             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1642             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1643             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1644
1645             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1646             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1647             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1648
1649             /**************************
1650              * CALCULATE INTERACTIONS *
1651              **************************/
1652
1653             r21              = _mm_mul_pd(rsq21,rinv21);
1654
1655             /* EWALD ELECTROSTATICS */
1656
1657             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1658             ewrt             = _mm_mul_pd(r21,ewtabscale);
1659             ewitab           = _mm_cvttpd_epi32(ewrt);
1660 #ifdef __XOP__
1661             eweps            = _mm_frcz_pd(ewrt);
1662 #else
1663             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1664 #endif
1665             twoeweps         = _mm_add_pd(eweps,eweps);
1666             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1667                                          &ewtabF,&ewtabFn);
1668             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1669             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1670
1671             fscal            = felec;
1672
1673             /* Update vectorial force */
1674             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1675             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1676             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1677
1678             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1679             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1680             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1681
1682             /**************************
1683              * CALCULATE INTERACTIONS *
1684              **************************/
1685
1686             r22              = _mm_mul_pd(rsq22,rinv22);
1687
1688             /* EWALD ELECTROSTATICS */
1689
1690             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1691             ewrt             = _mm_mul_pd(r22,ewtabscale);
1692             ewitab           = _mm_cvttpd_epi32(ewrt);
1693 #ifdef __XOP__
1694             eweps            = _mm_frcz_pd(ewrt);
1695 #else
1696             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1697 #endif
1698             twoeweps         = _mm_add_pd(eweps,eweps);
1699             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1700                                          &ewtabF,&ewtabFn);
1701             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1702             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1703
1704             fscal            = felec;
1705
1706             /* Update vectorial force */
1707             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1708             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1709             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1710
1711             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1712             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1713             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1714
1715             /**************************
1716              * CALCULATE INTERACTIONS *
1717              **************************/
1718
1719             r23              = _mm_mul_pd(rsq23,rinv23);
1720
1721             /* EWALD ELECTROSTATICS */
1722
1723             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1724             ewrt             = _mm_mul_pd(r23,ewtabscale);
1725             ewitab           = _mm_cvttpd_epi32(ewrt);
1726 #ifdef __XOP__
1727             eweps            = _mm_frcz_pd(ewrt);
1728 #else
1729             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1730 #endif
1731             twoeweps         = _mm_add_pd(eweps,eweps);
1732             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1733                                          &ewtabF,&ewtabFn);
1734             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1735             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1736
1737             fscal            = felec;
1738
1739             /* Update vectorial force */
1740             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1741             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1742             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1743
1744             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1745             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1746             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1747
1748             /**************************
1749              * CALCULATE INTERACTIONS *
1750              **************************/
1751
1752             r31              = _mm_mul_pd(rsq31,rinv31);
1753
1754             /* EWALD ELECTROSTATICS */
1755
1756             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1757             ewrt             = _mm_mul_pd(r31,ewtabscale);
1758             ewitab           = _mm_cvttpd_epi32(ewrt);
1759 #ifdef __XOP__
1760             eweps            = _mm_frcz_pd(ewrt);
1761 #else
1762             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1763 #endif
1764             twoeweps         = _mm_add_pd(eweps,eweps);
1765             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1766                                          &ewtabF,&ewtabFn);
1767             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1768             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1769
1770             fscal            = felec;
1771
1772             /* Update vectorial force */
1773             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1774             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1775             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1776
1777             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1778             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1779             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1780
1781             /**************************
1782              * CALCULATE INTERACTIONS *
1783              **************************/
1784
1785             r32              = _mm_mul_pd(rsq32,rinv32);
1786
1787             /* EWALD ELECTROSTATICS */
1788
1789             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1790             ewrt             = _mm_mul_pd(r32,ewtabscale);
1791             ewitab           = _mm_cvttpd_epi32(ewrt);
1792 #ifdef __XOP__
1793             eweps            = _mm_frcz_pd(ewrt);
1794 #else
1795             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1796 #endif
1797             twoeweps         = _mm_add_pd(eweps,eweps);
1798             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1799                                          &ewtabF,&ewtabFn);
1800             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1801             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1802
1803             fscal            = felec;
1804
1805             /* Update vectorial force */
1806             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1807             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1808             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1809
1810             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1811             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1812             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1813
1814             /**************************
1815              * CALCULATE INTERACTIONS *
1816              **************************/
1817
1818             r33              = _mm_mul_pd(rsq33,rinv33);
1819
1820             /* EWALD ELECTROSTATICS */
1821
1822             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1823             ewrt             = _mm_mul_pd(r33,ewtabscale);
1824             ewitab           = _mm_cvttpd_epi32(ewrt);
1825 #ifdef __XOP__
1826             eweps            = _mm_frcz_pd(ewrt);
1827 #else
1828             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1829 #endif
1830             twoeweps         = _mm_add_pd(eweps,eweps);
1831             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1832                                          &ewtabF,&ewtabFn);
1833             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1834             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1835
1836             fscal            = felec;
1837
1838             /* Update vectorial force */
1839             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1840             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1841             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1842
1843             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1844             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1845             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1846
1847             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);
1848
1849             /* Inner loop uses 384 flops */
1850         }
1851
1852         if(jidx<j_index_end)
1853         {
1854
1855             jnrA             = jjnr[jidx];
1856             j_coord_offsetA  = DIM*jnrA;
1857
1858             /* load j atom coordinates */
1859             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1860                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1861                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1862
1863             /* Calculate displacement vector */
1864             dx00             = _mm_sub_pd(ix0,jx0);
1865             dy00             = _mm_sub_pd(iy0,jy0);
1866             dz00             = _mm_sub_pd(iz0,jz0);
1867             dx11             = _mm_sub_pd(ix1,jx1);
1868             dy11             = _mm_sub_pd(iy1,jy1);
1869             dz11             = _mm_sub_pd(iz1,jz1);
1870             dx12             = _mm_sub_pd(ix1,jx2);
1871             dy12             = _mm_sub_pd(iy1,jy2);
1872             dz12             = _mm_sub_pd(iz1,jz2);
1873             dx13             = _mm_sub_pd(ix1,jx3);
1874             dy13             = _mm_sub_pd(iy1,jy3);
1875             dz13             = _mm_sub_pd(iz1,jz3);
1876             dx21             = _mm_sub_pd(ix2,jx1);
1877             dy21             = _mm_sub_pd(iy2,jy1);
1878             dz21             = _mm_sub_pd(iz2,jz1);
1879             dx22             = _mm_sub_pd(ix2,jx2);
1880             dy22             = _mm_sub_pd(iy2,jy2);
1881             dz22             = _mm_sub_pd(iz2,jz2);
1882             dx23             = _mm_sub_pd(ix2,jx3);
1883             dy23             = _mm_sub_pd(iy2,jy3);
1884             dz23             = _mm_sub_pd(iz2,jz3);
1885             dx31             = _mm_sub_pd(ix3,jx1);
1886             dy31             = _mm_sub_pd(iy3,jy1);
1887             dz31             = _mm_sub_pd(iz3,jz1);
1888             dx32             = _mm_sub_pd(ix3,jx2);
1889             dy32             = _mm_sub_pd(iy3,jy2);
1890             dz32             = _mm_sub_pd(iz3,jz2);
1891             dx33             = _mm_sub_pd(ix3,jx3);
1892             dy33             = _mm_sub_pd(iy3,jy3);
1893             dz33             = _mm_sub_pd(iz3,jz3);
1894
1895             /* Calculate squared distance and things based on it */
1896             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1897             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1898             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1899             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1900             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1901             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1902             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1903             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1904             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1905             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1906
1907             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1908             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1909             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1910             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1911             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1912             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1913             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1914             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1915             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1916
1917             rinvsq00         = gmx_mm_inv_pd(rsq00);
1918             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1919             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1920             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1921             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1922             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1923             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1924             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1925             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1926             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1927
1928             fjx0             = _mm_setzero_pd();
1929             fjy0             = _mm_setzero_pd();
1930             fjz0             = _mm_setzero_pd();
1931             fjx1             = _mm_setzero_pd();
1932             fjy1             = _mm_setzero_pd();
1933             fjz1             = _mm_setzero_pd();
1934             fjx2             = _mm_setzero_pd();
1935             fjy2             = _mm_setzero_pd();
1936             fjz2             = _mm_setzero_pd();
1937             fjx3             = _mm_setzero_pd();
1938             fjy3             = _mm_setzero_pd();
1939             fjz3             = _mm_setzero_pd();
1940
1941             /**************************
1942              * CALCULATE INTERACTIONS *
1943              **************************/
1944
1945             /* LENNARD-JONES DISPERSION/REPULSION */
1946
1947             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1948             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1949
1950             fscal            = fvdw;
1951
1952             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1953
1954             /* Update vectorial force */
1955             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1956             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1957             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1958
1959             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1960             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1961             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1962
1963             /**************************
1964              * CALCULATE INTERACTIONS *
1965              **************************/
1966
1967             r11              = _mm_mul_pd(rsq11,rinv11);
1968
1969             /* EWALD ELECTROSTATICS */
1970
1971             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1972             ewrt             = _mm_mul_pd(r11,ewtabscale);
1973             ewitab           = _mm_cvttpd_epi32(ewrt);
1974 #ifdef __XOP__
1975             eweps            = _mm_frcz_pd(ewrt);
1976 #else
1977             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1978 #endif
1979             twoeweps         = _mm_add_pd(eweps,eweps);
1980             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1981             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1982             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1983
1984             fscal            = felec;
1985
1986             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1987
1988             /* Update vectorial force */
1989             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1990             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1991             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1992
1993             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1994             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1995             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1996
1997             /**************************
1998              * CALCULATE INTERACTIONS *
1999              **************************/
2000
2001             r12              = _mm_mul_pd(rsq12,rinv12);
2002
2003             /* EWALD ELECTROSTATICS */
2004
2005             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2006             ewrt             = _mm_mul_pd(r12,ewtabscale);
2007             ewitab           = _mm_cvttpd_epi32(ewrt);
2008 #ifdef __XOP__
2009             eweps            = _mm_frcz_pd(ewrt);
2010 #else
2011             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2012 #endif
2013             twoeweps         = _mm_add_pd(eweps,eweps);
2014             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2015             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2016             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2017
2018             fscal            = felec;
2019
2020             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2021
2022             /* Update vectorial force */
2023             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2024             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2025             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2026
2027             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2028             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2029             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2030
2031             /**************************
2032              * CALCULATE INTERACTIONS *
2033              **************************/
2034
2035             r13              = _mm_mul_pd(rsq13,rinv13);
2036
2037             /* EWALD ELECTROSTATICS */
2038
2039             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2040             ewrt             = _mm_mul_pd(r13,ewtabscale);
2041             ewitab           = _mm_cvttpd_epi32(ewrt);
2042 #ifdef __XOP__
2043             eweps            = _mm_frcz_pd(ewrt);
2044 #else
2045             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2046 #endif
2047             twoeweps         = _mm_add_pd(eweps,eweps);
2048             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2049             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2050             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
2051
2052             fscal            = felec;
2053
2054             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2055
2056             /* Update vectorial force */
2057             fix1             = _mm_macc_pd(dx13,fscal,fix1);
2058             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
2059             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
2060
2061             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
2062             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
2063             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
2064
2065             /**************************
2066              * CALCULATE INTERACTIONS *
2067              **************************/
2068
2069             r21              = _mm_mul_pd(rsq21,rinv21);
2070
2071             /* EWALD ELECTROSTATICS */
2072
2073             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2074             ewrt             = _mm_mul_pd(r21,ewtabscale);
2075             ewitab           = _mm_cvttpd_epi32(ewrt);
2076 #ifdef __XOP__
2077             eweps            = _mm_frcz_pd(ewrt);
2078 #else
2079             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2080 #endif
2081             twoeweps         = _mm_add_pd(eweps,eweps);
2082             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2083             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2084             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2085
2086             fscal            = felec;
2087
2088             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2089
2090             /* Update vectorial force */
2091             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2092             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2093             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2094
2095             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2096             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2097             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2098
2099             /**************************
2100              * CALCULATE INTERACTIONS *
2101              **************************/
2102
2103             r22              = _mm_mul_pd(rsq22,rinv22);
2104
2105             /* EWALD ELECTROSTATICS */
2106
2107             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2108             ewrt             = _mm_mul_pd(r22,ewtabscale);
2109             ewitab           = _mm_cvttpd_epi32(ewrt);
2110 #ifdef __XOP__
2111             eweps            = _mm_frcz_pd(ewrt);
2112 #else
2113             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2114 #endif
2115             twoeweps         = _mm_add_pd(eweps,eweps);
2116             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2117             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2118             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2119
2120             fscal            = felec;
2121
2122             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2123
2124             /* Update vectorial force */
2125             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2126             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2127             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2128
2129             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2130             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2131             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2132
2133             /**************************
2134              * CALCULATE INTERACTIONS *
2135              **************************/
2136
2137             r23              = _mm_mul_pd(rsq23,rinv23);
2138
2139             /* EWALD ELECTROSTATICS */
2140
2141             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2142             ewrt             = _mm_mul_pd(r23,ewtabscale);
2143             ewitab           = _mm_cvttpd_epi32(ewrt);
2144 #ifdef __XOP__
2145             eweps            = _mm_frcz_pd(ewrt);
2146 #else
2147             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2148 #endif
2149             twoeweps         = _mm_add_pd(eweps,eweps);
2150             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2151             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2152             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
2153
2154             fscal            = felec;
2155
2156             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2157
2158             /* Update vectorial force */
2159             fix2             = _mm_macc_pd(dx23,fscal,fix2);
2160             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
2161             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
2162
2163             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
2164             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
2165             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
2166
2167             /**************************
2168              * CALCULATE INTERACTIONS *
2169              **************************/
2170
2171             r31              = _mm_mul_pd(rsq31,rinv31);
2172
2173             /* EWALD ELECTROSTATICS */
2174
2175             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2176             ewrt             = _mm_mul_pd(r31,ewtabscale);
2177             ewitab           = _mm_cvttpd_epi32(ewrt);
2178 #ifdef __XOP__
2179             eweps            = _mm_frcz_pd(ewrt);
2180 #else
2181             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2182 #endif
2183             twoeweps         = _mm_add_pd(eweps,eweps);
2184             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2185             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2186             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2187
2188             fscal            = felec;
2189
2190             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2191
2192             /* Update vectorial force */
2193             fix3             = _mm_macc_pd(dx31,fscal,fix3);
2194             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
2195             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
2196
2197             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
2198             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
2199             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
2200
2201             /**************************
2202              * CALCULATE INTERACTIONS *
2203              **************************/
2204
2205             r32              = _mm_mul_pd(rsq32,rinv32);
2206
2207             /* EWALD ELECTROSTATICS */
2208
2209             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2210             ewrt             = _mm_mul_pd(r32,ewtabscale);
2211             ewitab           = _mm_cvttpd_epi32(ewrt);
2212 #ifdef __XOP__
2213             eweps            = _mm_frcz_pd(ewrt);
2214 #else
2215             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2216 #endif
2217             twoeweps         = _mm_add_pd(eweps,eweps);
2218             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2219             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2220             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2221
2222             fscal            = felec;
2223
2224             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2225
2226             /* Update vectorial force */
2227             fix3             = _mm_macc_pd(dx32,fscal,fix3);
2228             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
2229             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
2230
2231             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
2232             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
2233             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
2234
2235             /**************************
2236              * CALCULATE INTERACTIONS *
2237              **************************/
2238
2239             r33              = _mm_mul_pd(rsq33,rinv33);
2240
2241             /* EWALD ELECTROSTATICS */
2242
2243             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2244             ewrt             = _mm_mul_pd(r33,ewtabscale);
2245             ewitab           = _mm_cvttpd_epi32(ewrt);
2246 #ifdef __XOP__
2247             eweps            = _mm_frcz_pd(ewrt);
2248 #else
2249             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2250 #endif
2251             twoeweps         = _mm_add_pd(eweps,eweps);
2252             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2253             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2254             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2255
2256             fscal            = felec;
2257
2258             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2259
2260             /* Update vectorial force */
2261             fix3             = _mm_macc_pd(dx33,fscal,fix3);
2262             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
2263             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
2264
2265             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
2266             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
2267             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
2268
2269             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2270
2271             /* Inner loop uses 384 flops */
2272         }
2273
2274         /* End of innermost loop */
2275
2276         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2277                                               f+i_coord_offset,fshift+i_shift_offset);
2278
2279         /* Increment number of inner iterations */
2280         inneriter                  += j_index_end - j_index_start;
2281
2282         /* Outer loop uses 24 flops */
2283     }
2284
2285     /* Increment number of outer iterations */
2286     outeriter        += nri;
2287
2288     /* Update outer/inner flops */
2289
2290     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*384);
2291 }