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