src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJEw_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4W4_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW4W4_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     int              vdwjidx1A,vdwjidx1B;
  93     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  94     int              vdwjidx2A,vdwjidx2B;
  95     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  96     int              vdwjidx3A,vdwjidx3B;
  97     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  98     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 100     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 101     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 102     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 103     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 104     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 105     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 106     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 107     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 108     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 109     real             *charge;
 110     int              nvdwtype;
 111     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 112     int              *vdwtype;
 113     real             *vdwparam;
 114     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 115     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 116     __m128d           c6grid_00;
 117     __m128d           c6grid_11;
 118     __m128d           c6grid_12;
 119     __m128d           c6grid_13;
 120     __m128d           c6grid_21;
 121     __m128d           c6grid_22;
 122     __m128d           c6grid_23;
 123     __m128d           c6grid_31;
 124     __m128d           c6grid_32;
 125     __m128d           c6grid_33;
 126     real             *vdwgridparam;
 127     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 128     __m128d           one_half  = _mm_set1_pd(0.5);
 129     __m128d           minus_one = _mm_set1_pd(-1.0);
 130     __m128i          ewitab;
 131     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 132     real             *ewtab;
 133     __m128d          dummy_mask,cutoff_mask;
 134     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 135     __m128d          one     = _mm_set1_pd(1.0);
 136     __m128d          two     = _mm_set1_pd(2.0);
 137     x                = xx[0];
 138     f                = ff[0];
 139
 140     nri              = nlist->nri;
 141     iinr             = nlist->iinr;
 142     jindex           = nlist->jindex;
 143     jjnr             = nlist->jjnr;
 144     shiftidx         = nlist->shift;
 145     gid              = nlist->gid;
 146     shiftvec         = fr->shift_vec[0];
 147     fshift           = fr->fshift[0];
 148     facel            = _mm_set1_pd(fr->epsfac);
 149     charge           = mdatoms->chargeA;
 150     nvdwtype         = fr->ntype;
 151     vdwparam         = fr->nbfp;
 152     vdwtype          = mdatoms->typeA;
 153     vdwgridparam     = fr->ljpme_c6grid;
 154     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 155     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 156     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 157
 158     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 159     ewtab            = fr->ic->tabq_coul_FDV0;
 160     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 161     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 162
 163     /* Setup water-specific parameters */
 164     inr              = nlist->iinr[0];
 165     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 166     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 167     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 168     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 169
 170     jq1              = _mm_set1_pd(charge[inr+1]);
 171     jq2              = _mm_set1_pd(charge[inr+2]);
 172     jq3              = _mm_set1_pd(charge[inr+3]);
 173     vdwjidx0A        = 2*vdwtype[inr+0];
 174     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 175     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 176     c6grid_00        = _mm_set1_pd(vdwgridparam[vdwioffset0+vdwjidx0A]);
 177     qq11             = _mm_mul_pd(iq1,jq1);
 178     qq12             = _mm_mul_pd(iq1,jq2);
 179     qq13             = _mm_mul_pd(iq1,jq3);
 180     qq21             = _mm_mul_pd(iq2,jq1);
 181     qq22             = _mm_mul_pd(iq2,jq2);
 182     qq23             = _mm_mul_pd(iq2,jq3);
 183     qq31             = _mm_mul_pd(iq3,jq1);
 184     qq32             = _mm_mul_pd(iq3,jq2);
 185     qq33             = _mm_mul_pd(iq3,jq3);
 186
 187     /* Avoid stupid compiler warnings */
 188     jnrA = jnrB = 0;
 189     j_coord_offsetA = 0;
 190     j_coord_offsetB = 0;
 191
 192     outeriter        = 0;
 193     inneriter        = 0;
 194
 195     /* Start outer loop over neighborlists */
 196     for(iidx=0; iidx<nri; iidx++)
 197     {
 198         /* Load shift vector for this list */
 199         i_shift_offset   = DIM*shiftidx[iidx];
 200
 201         /* Load limits for loop over neighbors */
 202         j_index_start    = jindex[iidx];
 203         j_index_end      = jindex[iidx+1];
 204
 205         /* Get outer coordinate index */
 206         inr              = iinr[iidx];
 207         i_coord_offset   = DIM*inr;
 208
 209         /* Load i particle coords and add shift vector */
 210         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 211                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 212
 213         fix0             = _mm_setzero_pd();
 214         fiy0             = _mm_setzero_pd();
 215         fiz0             = _mm_setzero_pd();
 216         fix1             = _mm_setzero_pd();
 217         fiy1             = _mm_setzero_pd();
 218         fiz1             = _mm_setzero_pd();
 219         fix2             = _mm_setzero_pd();
 220         fiy2             = _mm_setzero_pd();
 221         fiz2             = _mm_setzero_pd();
 222         fix3             = _mm_setzero_pd();
 223         fiy3             = _mm_setzero_pd();
 224         fiz3             = _mm_setzero_pd();
 225
 226         /* Reset potential sums */
 227         velecsum         = _mm_setzero_pd();
 228         vvdwsum          = _mm_setzero_pd();
 229
 230         /* Start inner kernel loop */
 231         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 232         {
 233
 234             /* Get j neighbor index, and coordinate index */
 235             jnrA             = jjnr[jidx];
 236             jnrB             = jjnr[jidx+1];
 237             j_coord_offsetA  = DIM*jnrA;
 238             j_coord_offsetB  = DIM*jnrB;
 239
 240             /* load j atom coordinates */
 241             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 242                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 243                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 244
 245             /* Calculate displacement vector */
 246             dx00             = _mm_sub_pd(ix0,jx0);
 247             dy00             = _mm_sub_pd(iy0,jy0);
 248             dz00             = _mm_sub_pd(iz0,jz0);
 249             dx11             = _mm_sub_pd(ix1,jx1);
 250             dy11             = _mm_sub_pd(iy1,jy1);
 251             dz11             = _mm_sub_pd(iz1,jz1);
 252             dx12             = _mm_sub_pd(ix1,jx2);
 253             dy12             = _mm_sub_pd(iy1,jy2);
 254             dz12             = _mm_sub_pd(iz1,jz2);
 255             dx13             = _mm_sub_pd(ix1,jx3);
 256             dy13             = _mm_sub_pd(iy1,jy3);
 257             dz13             = _mm_sub_pd(iz1,jz3);
 258             dx21             = _mm_sub_pd(ix2,jx1);
 259             dy21             = _mm_sub_pd(iy2,jy1);
 260             dz21             = _mm_sub_pd(iz2,jz1);
 261             dx22             = _mm_sub_pd(ix2,jx2);
 262             dy22             = _mm_sub_pd(iy2,jy2);
 263             dz22             = _mm_sub_pd(iz2,jz2);
 264             dx23             = _mm_sub_pd(ix2,jx3);
 265             dy23             = _mm_sub_pd(iy2,jy3);
 266             dz23             = _mm_sub_pd(iz2,jz3);
 267             dx31             = _mm_sub_pd(ix3,jx1);
 268             dy31             = _mm_sub_pd(iy3,jy1);
 269             dz31             = _mm_sub_pd(iz3,jz1);
 270             dx32             = _mm_sub_pd(ix3,jx2);
 271             dy32             = _mm_sub_pd(iy3,jy2);
 272             dz32             = _mm_sub_pd(iz3,jz2);
 273             dx33             = _mm_sub_pd(ix3,jx3);
 274             dy33             = _mm_sub_pd(iy3,jy3);
 275             dz33             = _mm_sub_pd(iz3,jz3);
 276
 277             /* Calculate squared distance and things based on it */
 278             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 279             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 280             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 281             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 282             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 283             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 284             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 285             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 286             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 287             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 288
 289             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 290             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 291             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 292             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 293             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 294             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 295             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 296             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 297             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 298             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 299
 300             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 301             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 302             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 303             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 304             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 305             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 306             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 307             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 308             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 309             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 310
 311             fjx0             = _mm_setzero_pd();
 312             fjy0             = _mm_setzero_pd();
 313             fjz0             = _mm_setzero_pd();
 314             fjx1             = _mm_setzero_pd();
 315             fjy1             = _mm_setzero_pd();
 316             fjz1             = _mm_setzero_pd();
 317             fjx2             = _mm_setzero_pd();
 318             fjy2             = _mm_setzero_pd();
 319             fjz2             = _mm_setzero_pd();
 320             fjx3             = _mm_setzero_pd();
 321             fjy3             = _mm_setzero_pd();
 322             fjz3             = _mm_setzero_pd();
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             r00              = _mm_mul_pd(rsq00,rinv00);
 329
 330             /* Analytical LJ-PME */
 331             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 332             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 333             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 334             exponent         = gmx_simd_exp_d(ewcljrsq);
 335             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 336             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 337             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 338             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 339             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 340             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 341             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 342             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 346
 347             fscal            = fvdw;
 348
 349             /* Update vectorial force */
 350             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 351             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 352             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 353
 354             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 355             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 356             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             r11              = _mm_mul_pd(rsq11,rinv11);
 363
 364             /* EWALD ELECTROSTATICS */
 365
 366             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 367             ewrt             = _mm_mul_pd(r11,ewtabscale);
 368             ewitab           = _mm_cvttpd_epi32(ewrt);
 369 #ifdef __XOP__
 370             eweps            = _mm_frcz_pd(ewrt);
 371 #else
 372             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 373 #endif
 374             twoeweps         = _mm_add_pd(eweps,eweps);
 375             ewitab           = _mm_slli_epi32(ewitab,2);
 376             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 377             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 378             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 379             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 380             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 381             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 382             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 383             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 384             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 385             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velecsum         = _mm_add_pd(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             /* Update vectorial force */
 393             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 394             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 395             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 396
 397             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 398             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 399             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             r12              = _mm_mul_pd(rsq12,rinv12);
 406
 407             /* EWALD ELECTROSTATICS */
 408
 409             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 410             ewrt             = _mm_mul_pd(r12,ewtabscale);
 411             ewitab           = _mm_cvttpd_epi32(ewrt);
 412 #ifdef __XOP__
 413             eweps            = _mm_frcz_pd(ewrt);
 414 #else
 415             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 416 #endif
 417             twoeweps         = _mm_add_pd(eweps,eweps);
 418             ewitab           = _mm_slli_epi32(ewitab,2);
 419             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 420             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 421             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 422             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 423             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 424             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 425             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 426             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 427             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 428             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velecsum         = _mm_add_pd(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435             /* Update vectorial force */
 436             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 437             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 438             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 439
 440             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 441             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 442             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             r13              = _mm_mul_pd(rsq13,rinv13);
 449
 450             /* EWALD ELECTROSTATICS */
 451
 452             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 453             ewrt             = _mm_mul_pd(r13,ewtabscale);
 454             ewitab           = _mm_cvttpd_epi32(ewrt);
 455 #ifdef __XOP__
 456             eweps            = _mm_frcz_pd(ewrt);
 457 #else
 458             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 459 #endif
 460             twoeweps         = _mm_add_pd(eweps,eweps);
 461             ewitab           = _mm_slli_epi32(ewitab,2);
 462             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 463             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 464             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 465             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 466             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 467             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 468             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 469             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 470             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
 471             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 472
 473             /* Update potential sum for this i atom from the interaction with this j atom. */
 474             velecsum         = _mm_add_pd(velecsum,velec);
 475
 476             fscal            = felec;
 477
 478             /* Update vectorial force */
 479             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 480             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 481             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 482
 483             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 484             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 485             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             r21              = _mm_mul_pd(rsq21,rinv21);
 492
 493             /* EWALD ELECTROSTATICS */
 494
 495             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 496             ewrt             = _mm_mul_pd(r21,ewtabscale);
 497             ewitab           = _mm_cvttpd_epi32(ewrt);
 498 #ifdef __XOP__
 499             eweps            = _mm_frcz_pd(ewrt);
 500 #else
 501             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 502 #endif
 503             twoeweps         = _mm_add_pd(eweps,eweps);
 504             ewitab           = _mm_slli_epi32(ewitab,2);
 505             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 506             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 507             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 508             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 509             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 510             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 511             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 512             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 513             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 514             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velecsum         = _mm_add_pd(velecsum,velec);
 518
 519             fscal            = felec;
 520
 521             /* Update vectorial force */
 522             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 523             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 524             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 525
 526             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 527             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 528             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             r22              = _mm_mul_pd(rsq22,rinv22);
 535
 536             /* EWALD ELECTROSTATICS */
 537
 538             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 539             ewrt             = _mm_mul_pd(r22,ewtabscale);
 540             ewitab           = _mm_cvttpd_epi32(ewrt);
 541 #ifdef __XOP__
 542             eweps            = _mm_frcz_pd(ewrt);
 543 #else
 544             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 545 #endif
 546             twoeweps         = _mm_add_pd(eweps,eweps);
 547             ewitab           = _mm_slli_epi32(ewitab,2);
 548             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 549             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 550             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 551             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 552             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 553             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 554             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 555             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 556             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 557             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 558
 559             /* Update potential sum for this i atom from the interaction with this j atom. */
 560             velecsum         = _mm_add_pd(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             /* Update vectorial force */
 565             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 566             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 567             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 568
 569             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 570             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 571             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             r23              = _mm_mul_pd(rsq23,rinv23);
 578
 579             /* EWALD ELECTROSTATICS */
 580
 581             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 582             ewrt             = _mm_mul_pd(r23,ewtabscale);
 583             ewitab           = _mm_cvttpd_epi32(ewrt);
 584 #ifdef __XOP__
 585             eweps            = _mm_frcz_pd(ewrt);
 586 #else
 587             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 588 #endif
 589             twoeweps         = _mm_add_pd(eweps,eweps);
 590             ewitab           = _mm_slli_epi32(ewitab,2);
 591             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 592             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 593             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 594             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 595             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 596             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 597             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 598             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 599             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
 600             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
 601
 602             /* Update potential sum for this i atom from the interaction with this j atom. */
 603             velecsum         = _mm_add_pd(velecsum,velec);
 604
 605             fscal            = felec;
 606
 607             /* Update vectorial force */
 608             fix2             = _mm_macc_pd(dx23,fscal,fix2);
 609             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
 610             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
 611
 612             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
 613             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
 614             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             r31              = _mm_mul_pd(rsq31,rinv31);
 621
 622             /* EWALD ELECTROSTATICS */
 623
 624             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 625             ewrt             = _mm_mul_pd(r31,ewtabscale);
 626             ewitab           = _mm_cvttpd_epi32(ewrt);
 627 #ifdef __XOP__
 628             eweps            = _mm_frcz_pd(ewrt);
 629 #else
 630             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 631 #endif
 632             twoeweps         = _mm_add_pd(eweps,eweps);
 633             ewitab           = _mm_slli_epi32(ewitab,2);
 634             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 635             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 636             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 637             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 638             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 639             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 640             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 641             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 642             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
 643             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
 644
 645             /* Update potential sum for this i atom from the interaction with this j atom. */
 646             velecsum         = _mm_add_pd(velecsum,velec);
 647
 648             fscal            = felec;
 649
 650             /* Update vectorial force */
 651             fix3             = _mm_macc_pd(dx31,fscal,fix3);
 652             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
 653             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
 654
 655             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
 656             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
 657             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             r32              = _mm_mul_pd(rsq32,rinv32);
 664
 665             /* EWALD ELECTROSTATICS */
 666
 667             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 668             ewrt             = _mm_mul_pd(r32,ewtabscale);
 669             ewitab           = _mm_cvttpd_epi32(ewrt);
 670 #ifdef __XOP__
 671             eweps            = _mm_frcz_pd(ewrt);
 672 #else
 673             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 674 #endif
 675             twoeweps         = _mm_add_pd(eweps,eweps);
 676             ewitab           = _mm_slli_epi32(ewitab,2);
 677             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 678             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 679             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 680             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 681             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 682             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 683             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 684             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 685             velec            = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
 686             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
 687
 688             /* Update potential sum for this i atom from the interaction with this j atom. */
 689             velecsum         = _mm_add_pd(velecsum,velec);
 690
 691             fscal            = felec;
 692
 693             /* Update vectorial force */
 694             fix3             = _mm_macc_pd(dx32,fscal,fix3);
 695             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
 696             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
 697
 698             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
 699             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
 700             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
 701
 702             /**************************
 703              * CALCULATE INTERACTIONS *
 704              **************************/
 705
 706             r33              = _mm_mul_pd(rsq33,rinv33);
 707
 708             /* EWALD ELECTROSTATICS */
 709
 710             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 711             ewrt             = _mm_mul_pd(r33,ewtabscale);
 712             ewitab           = _mm_cvttpd_epi32(ewrt);
 713 #ifdef __XOP__
 714             eweps            = _mm_frcz_pd(ewrt);
 715 #else
 716             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 717 #endif
 718             twoeweps         = _mm_add_pd(eweps,eweps);
 719             ewitab           = _mm_slli_epi32(ewitab,2);
 720             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 721             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 722             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 723             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 724             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 725             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 726             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 727             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 728             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
 729             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
 730
 731             /* Update potential sum for this i atom from the interaction with this j atom. */
 732             velecsum         = _mm_add_pd(velecsum,velec);
 733
 734             fscal            = felec;
 735
 736             /* Update vectorial force */
 737             fix3             = _mm_macc_pd(dx33,fscal,fix3);
 738             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
 739             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
 740
 741             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
 742             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
 743             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
 744
 745             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);
 746
 747             /* Inner loop uses 449 flops */
 748         }
 749
 750         if(jidx<j_index_end)
 751         {
 752
 753             jnrA             = jjnr[jidx];
 754             j_coord_offsetA  = DIM*jnrA;
 755
 756             /* load j atom coordinates */
 757             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 758                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 759                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 760
 761             /* Calculate displacement vector */
 762             dx00             = _mm_sub_pd(ix0,jx0);
 763             dy00             = _mm_sub_pd(iy0,jy0);
 764             dz00             = _mm_sub_pd(iz0,jz0);
 765             dx11             = _mm_sub_pd(ix1,jx1);
 766             dy11             = _mm_sub_pd(iy1,jy1);
 767             dz11             = _mm_sub_pd(iz1,jz1);
 768             dx12             = _mm_sub_pd(ix1,jx2);
 769             dy12             = _mm_sub_pd(iy1,jy2);
 770             dz12             = _mm_sub_pd(iz1,jz2);
 771             dx13             = _mm_sub_pd(ix1,jx3);
 772             dy13             = _mm_sub_pd(iy1,jy3);
 773             dz13             = _mm_sub_pd(iz1,jz3);
 774             dx21             = _mm_sub_pd(ix2,jx1);
 775             dy21             = _mm_sub_pd(iy2,jy1);
 776             dz21             = _mm_sub_pd(iz2,jz1);
 777             dx22             = _mm_sub_pd(ix2,jx2);
 778             dy22             = _mm_sub_pd(iy2,jy2);
 779             dz22             = _mm_sub_pd(iz2,jz2);
 780             dx23             = _mm_sub_pd(ix2,jx3);
 781             dy23             = _mm_sub_pd(iy2,jy3);
 782             dz23             = _mm_sub_pd(iz2,jz3);
 783             dx31             = _mm_sub_pd(ix3,jx1);
 784             dy31             = _mm_sub_pd(iy3,jy1);
 785             dz31             = _mm_sub_pd(iz3,jz1);
 786             dx32             = _mm_sub_pd(ix3,jx2);
 787             dy32             = _mm_sub_pd(iy3,jy2);
 788             dz32             = _mm_sub_pd(iz3,jz2);
 789             dx33             = _mm_sub_pd(ix3,jx3);
 790             dy33             = _mm_sub_pd(iy3,jy3);
 791             dz33             = _mm_sub_pd(iz3,jz3);
 792
 793             /* Calculate squared distance and things based on it */
 794             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 795             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 796             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 797             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 798             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 799             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 800             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 801             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 802             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 803             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 804
 805             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 806             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 807             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 808             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 809             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 810             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 811             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 812             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 813             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 814             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 815
 816             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 817             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 818             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 819             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 820             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 821             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 822             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 823             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 824             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 825             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 826
 827             fjx0             = _mm_setzero_pd();
 828             fjy0             = _mm_setzero_pd();
 829             fjz0             = _mm_setzero_pd();
 830             fjx1             = _mm_setzero_pd();
 831             fjy1             = _mm_setzero_pd();
 832             fjz1             = _mm_setzero_pd();
 833             fjx2             = _mm_setzero_pd();
 834             fjy2             = _mm_setzero_pd();
 835             fjz2             = _mm_setzero_pd();
 836             fjx3             = _mm_setzero_pd();
 837             fjy3             = _mm_setzero_pd();
 838             fjz3             = _mm_setzero_pd();
 839
 840             /**************************
 841              * CALCULATE INTERACTIONS *
 842              **************************/
 843
 844             r00              = _mm_mul_pd(rsq00,rinv00);
 845
 846             /* Analytical LJ-PME */
 847             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 848             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 849             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 850             exponent         = gmx_simd_exp_d(ewcljrsq);
 851             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 852             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 853             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 854             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 855             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 856             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 857             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 858             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 859
 860             /* Update potential sum for this i atom from the interaction with this j atom. */
 861             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 862             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 863
 864             fscal            = fvdw;
 865
 866             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 867
 868             /* Update vectorial force */
 869             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 870             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 871             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 872
 873             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 874             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 875             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 876
 877             /**************************
 878              * CALCULATE INTERACTIONS *
 879              **************************/
 880
 881             r11              = _mm_mul_pd(rsq11,rinv11);
 882
 883             /* EWALD ELECTROSTATICS */
 884
 885             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 886             ewrt             = _mm_mul_pd(r11,ewtabscale);
 887             ewitab           = _mm_cvttpd_epi32(ewrt);
 888 #ifdef __XOP__
 889             eweps            = _mm_frcz_pd(ewrt);
 890 #else
 891             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 892 #endif
 893             twoeweps         = _mm_add_pd(eweps,eweps);
 894             ewitab           = _mm_slli_epi32(ewitab,2);
 895             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 896             ewtabD           = _mm_setzero_pd();
 897             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 898             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 899             ewtabFn          = _mm_setzero_pd();
 900             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 901             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 902             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 903             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 904             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 905
 906             /* Update potential sum for this i atom from the interaction with this j atom. */
 907             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 908             velecsum         = _mm_add_pd(velecsum,velec);
 909
 910             fscal            = felec;
 911
 912             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 913
 914             /* Update vectorial force */
 915             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 916             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 917             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 918
 919             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 920             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 921             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 922
 923             /**************************
 924              * CALCULATE INTERACTIONS *
 925              **************************/
 926
 927             r12              = _mm_mul_pd(rsq12,rinv12);
 928
 929             /* EWALD ELECTROSTATICS */
 930
 931             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 932             ewrt             = _mm_mul_pd(r12,ewtabscale);
 933             ewitab           = _mm_cvttpd_epi32(ewrt);
 934 #ifdef __XOP__
 935             eweps            = _mm_frcz_pd(ewrt);
 936 #else
 937             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 938 #endif
 939             twoeweps         = _mm_add_pd(eweps,eweps);
 940             ewitab           = _mm_slli_epi32(ewitab,2);
 941             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 942             ewtabD           = _mm_setzero_pd();
 943             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 944             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 945             ewtabFn          = _mm_setzero_pd();
 946             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 947             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 948             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 949             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 950             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 951
 952             /* Update potential sum for this i atom from the interaction with this j atom. */
 953             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 954             velecsum         = _mm_add_pd(velecsum,velec);
 955
 956             fscal            = felec;
 957
 958             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 959
 960             /* Update vectorial force */
 961             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 962             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 963             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 964
 965             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 966             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 967             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             r13              = _mm_mul_pd(rsq13,rinv13);
 974
 975             /* EWALD ELECTROSTATICS */
 976
 977             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 978             ewrt             = _mm_mul_pd(r13,ewtabscale);
 979             ewitab           = _mm_cvttpd_epi32(ewrt);
 980 #ifdef __XOP__
 981             eweps            = _mm_frcz_pd(ewrt);
 982 #else
 983             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 984 #endif
 985             twoeweps         = _mm_add_pd(eweps,eweps);
 986             ewitab           = _mm_slli_epi32(ewitab,2);
 987             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 988             ewtabD           = _mm_setzero_pd();
 989             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 990             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 991             ewtabFn          = _mm_setzero_pd();
 992             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 993             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 994             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 995             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
 996             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 997
 998             /* Update potential sum for this i atom from the interaction with this j atom. */
 999             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1000             velecsum         = _mm_add_pd(velecsum,velec);
1001
1002             fscal            = felec;
1003
1004             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1005
1006             /* Update vectorial force */
1007             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1008             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1009             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1010
1011             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1012             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1013             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             r21              = _mm_mul_pd(rsq21,rinv21);
1020
1021             /* EWALD ELECTROSTATICS */
1022
1023             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1024             ewrt             = _mm_mul_pd(r21,ewtabscale);
1025             ewitab           = _mm_cvttpd_epi32(ewrt);
1026 #ifdef __XOP__
1027             eweps            = _mm_frcz_pd(ewrt);
1028 #else
1029             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1030 #endif
1031             twoeweps         = _mm_add_pd(eweps,eweps);
1032             ewitab           = _mm_slli_epi32(ewitab,2);
1033             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1034             ewtabD           = _mm_setzero_pd();
1035             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1036             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1037             ewtabFn          = _mm_setzero_pd();
1038             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1039             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1040             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1041             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1042             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1043
1044             /* Update potential sum for this i atom from the interaction with this j atom. */
1045             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1046             velecsum         = _mm_add_pd(velecsum,velec);
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1051
1052             /* Update vectorial force */
1053             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1054             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1055             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1056
1057             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1058             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1059             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1060
1061             /**************************
1062              * CALCULATE INTERACTIONS *
1063              **************************/
1064
1065             r22              = _mm_mul_pd(rsq22,rinv22);
1066
1067             /* EWALD ELECTROSTATICS */
1068
1069             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1070             ewrt             = _mm_mul_pd(r22,ewtabscale);
1071             ewitab           = _mm_cvttpd_epi32(ewrt);
1072 #ifdef __XOP__
1073             eweps            = _mm_frcz_pd(ewrt);
1074 #else
1075             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1076 #endif
1077             twoeweps         = _mm_add_pd(eweps,eweps);
1078             ewitab           = _mm_slli_epi32(ewitab,2);
1079             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1080             ewtabD           = _mm_setzero_pd();
1081             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1082             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1083             ewtabFn          = _mm_setzero_pd();
1084             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1085             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1086             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1087             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1088             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1089
1090             /* Update potential sum for this i atom from the interaction with this j atom. */
1091             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1092             velecsum         = _mm_add_pd(velecsum,velec);
1093
1094             fscal            = felec;
1095
1096             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1097
1098             /* Update vectorial force */
1099             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1100             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1101             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1102
1103             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1104             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1105             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1106
1107             /**************************
1108              * CALCULATE INTERACTIONS *
1109              **************************/
1110
1111             r23              = _mm_mul_pd(rsq23,rinv23);
1112
1113             /* EWALD ELECTROSTATICS */
1114
1115             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1116             ewrt             = _mm_mul_pd(r23,ewtabscale);
1117             ewitab           = _mm_cvttpd_epi32(ewrt);
1118 #ifdef __XOP__
1119             eweps            = _mm_frcz_pd(ewrt);
1120 #else
1121             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1122 #endif
1123             twoeweps         = _mm_add_pd(eweps,eweps);
1124             ewitab           = _mm_slli_epi32(ewitab,2);
1125             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1126             ewtabD           = _mm_setzero_pd();
1127             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1128             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1129             ewtabFn          = _mm_setzero_pd();
1130             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1131             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1132             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1133             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
1134             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1135
1136             /* Update potential sum for this i atom from the interaction with this j atom. */
1137             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1138             velecsum         = _mm_add_pd(velecsum,velec);
1139
1140             fscal            = felec;
1141
1142             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1143
1144             /* Update vectorial force */
1145             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1146             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1147             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1148
1149             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1150             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1151             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             r31              = _mm_mul_pd(rsq31,rinv31);
1158
1159             /* EWALD ELECTROSTATICS */
1160
1161             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1162             ewrt             = _mm_mul_pd(r31,ewtabscale);
1163             ewitab           = _mm_cvttpd_epi32(ewrt);
1164 #ifdef __XOP__
1165             eweps            = _mm_frcz_pd(ewrt);
1166 #else
1167             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1168 #endif
1169             twoeweps         = _mm_add_pd(eweps,eweps);
1170             ewitab           = _mm_slli_epi32(ewitab,2);
1171             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1172             ewtabD           = _mm_setzero_pd();
1173             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1174             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1175             ewtabFn          = _mm_setzero_pd();
1176             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1177             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1178             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1179             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
1180             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1181
1182             /* Update potential sum for this i atom from the interaction with this j atom. */
1183             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1184             velecsum         = _mm_add_pd(velecsum,velec);
1185
1186             fscal            = felec;
1187
1188             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1189
1190             /* Update vectorial force */
1191             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1192             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1193             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1194
1195             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1196             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1197             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             r32              = _mm_mul_pd(rsq32,rinv32);
1204
1205             /* EWALD ELECTROSTATICS */
1206
1207             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1208             ewrt             = _mm_mul_pd(r32,ewtabscale);
1209             ewitab           = _mm_cvttpd_epi32(ewrt);
1210 #ifdef __XOP__
1211             eweps            = _mm_frcz_pd(ewrt);
1212 #else
1213             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1214 #endif
1215             twoeweps         = _mm_add_pd(eweps,eweps);
1216             ewitab           = _mm_slli_epi32(ewitab,2);
1217             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1218             ewtabD           = _mm_setzero_pd();
1219             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1220             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1221             ewtabFn          = _mm_setzero_pd();
1222             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1223             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1224             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1225             velec            = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
1226             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1227
1228             /* Update potential sum for this i atom from the interaction with this j atom. */
1229             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1230             velecsum         = _mm_add_pd(velecsum,velec);
1231
1232             fscal            = felec;
1233
1234             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1235
1236             /* Update vectorial force */
1237             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1238             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1239             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1240
1241             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1242             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1243             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1244
1245             /**************************
1246              * CALCULATE INTERACTIONS *
1247              **************************/
1248
1249             r33              = _mm_mul_pd(rsq33,rinv33);
1250
1251             /* EWALD ELECTROSTATICS */
1252
1253             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1254             ewrt             = _mm_mul_pd(r33,ewtabscale);
1255             ewitab           = _mm_cvttpd_epi32(ewrt);
1256 #ifdef __XOP__
1257             eweps            = _mm_frcz_pd(ewrt);
1258 #else
1259             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1260 #endif
1261             twoeweps         = _mm_add_pd(eweps,eweps);
1262             ewitab           = _mm_slli_epi32(ewitab,2);
1263             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1264             ewtabD           = _mm_setzero_pd();
1265             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1266             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1267             ewtabFn          = _mm_setzero_pd();
1268             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1269             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1270             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1271             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
1272             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1273
1274             /* Update potential sum for this i atom from the interaction with this j atom. */
1275             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1276             velecsum         = _mm_add_pd(velecsum,velec);
1277
1278             fscal            = felec;
1279
1280             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1281
1282             /* Update vectorial force */
1283             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1284             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1285             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1286
1287             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1288             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1289             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1290
1291             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1292
1293             /* Inner loop uses 449 flops */
1294         }
1295
1296         /* End of innermost loop */
1297
1298         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1299                                               f+i_coord_offset,fshift+i_shift_offset);
1300
1301         ggid                        = gid[iidx];
1302         /* Update potential energies */
1303         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1304         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1305
1306         /* Increment number of inner iterations */
1307         inneriter                  += j_index_end - j_index_start;
1308
1309         /* Outer loop uses 26 flops */
1310     }
1311
1312     /* Increment number of outer iterations */
1313     outeriter        += nri;
1314
1315     /* Update outer/inner flops */
1316
1317     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*449);
1318 }
1319 /*
1320  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4W4_F_avx_128_fma_double
1321  * Electrostatics interaction: Ewald
1322  * VdW interaction:            LJEwald
1323  * Geometry:                   Water4-Water4
1324  * Calculate force/pot:        Force
1325  */
1326 void
1327 nb_kernel_ElecEw_VdwLJEw_GeomW4W4_F_avx_128_fma_double
1328                     (t_nblist                    * gmx_restrict       nlist,
1329                      rvec                        * gmx_restrict          xx,
1330                      rvec                        * gmx_restrict          ff,
1331                      t_forcerec                  * gmx_restrict          fr,
1332                      t_mdatoms                   * gmx_restrict     mdatoms,
1333                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1334                      t_nrnb                      * gmx_restrict        nrnb)
1335 {
1336     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1337      * just 0 for non-waters.
1338      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1339      * jnr indices corresponding to data put in the four positions in the SIMD register.
1340      */
1341     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1342     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1343     int              jnrA,jnrB;
1344     int              j_coord_offsetA,j_coord_offsetB;
1345     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1346     real             rcutoff_scalar;
1347     real             *shiftvec,*fshift,*x,*f;
1348     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1349     int              vdwioffset0;
1350     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1351     int              vdwioffset1;
1352     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1353     int              vdwioffset2;
1354     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1355     int              vdwioffset3;
1356     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1357     int              vdwjidx0A,vdwjidx0B;
1358     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1359     int              vdwjidx1A,vdwjidx1B;
1360     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1361     int              vdwjidx2A,vdwjidx2B;
1362     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1363     int              vdwjidx3A,vdwjidx3B;
1364     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1365     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1366     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1367     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1368     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1369     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1370     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1371     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1372     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1373     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1374     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1375     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1376     real             *charge;
1377     int              nvdwtype;
1378     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1379     int              *vdwtype;
1380     real             *vdwparam;
1381     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1382     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1383     __m128d           c6grid_00;
1384     __m128d           c6grid_11;
1385     __m128d           c6grid_12;
1386     __m128d           c6grid_13;
1387     __m128d           c6grid_21;
1388     __m128d           c6grid_22;
1389     __m128d           c6grid_23;
1390     __m128d           c6grid_31;
1391     __m128d           c6grid_32;
1392     __m128d           c6grid_33;
1393     real             *vdwgridparam;
1394     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1395     __m128d           one_half  = _mm_set1_pd(0.5);
1396     __m128d           minus_one = _mm_set1_pd(-1.0);
1397     __m128i          ewitab;
1398     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1399     real             *ewtab;
1400     __m128d          dummy_mask,cutoff_mask;
1401     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1402     __m128d          one     = _mm_set1_pd(1.0);
1403     __m128d          two     = _mm_set1_pd(2.0);
1404     x                = xx[0];
1405     f                = ff[0];
1406
1407     nri              = nlist->nri;
1408     iinr             = nlist->iinr;
1409     jindex           = nlist->jindex;
1410     jjnr             = nlist->jjnr;
1411     shiftidx         = nlist->shift;
1412     gid              = nlist->gid;
1413     shiftvec         = fr->shift_vec[0];
1414     fshift           = fr->fshift[0];
1415     facel            = _mm_set1_pd(fr->epsfac);
1416     charge           = mdatoms->chargeA;
1417     nvdwtype         = fr->ntype;
1418     vdwparam         = fr->nbfp;
1419     vdwtype          = mdatoms->typeA;
1420     vdwgridparam     = fr->ljpme_c6grid;
1421     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
1422     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
1423     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
1424
1425     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1426     ewtab            = fr->ic->tabq_coul_F;
1427     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1428     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1429
1430     /* Setup water-specific parameters */
1431     inr              = nlist->iinr[0];
1432     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1433     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1434     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1435     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1436
1437     jq1              = _mm_set1_pd(charge[inr+1]);
1438     jq2              = _mm_set1_pd(charge[inr+2]);
1439     jq3              = _mm_set1_pd(charge[inr+3]);
1440     vdwjidx0A        = 2*vdwtype[inr+0];
1441     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1442     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1443     c6grid_00        = _mm_set1_pd(vdwgridparam[vdwioffset0+vdwjidx0A]);
1444     qq11             = _mm_mul_pd(iq1,jq1);
1445     qq12             = _mm_mul_pd(iq1,jq2);
1446     qq13             = _mm_mul_pd(iq1,jq3);
1447     qq21             = _mm_mul_pd(iq2,jq1);
1448     qq22             = _mm_mul_pd(iq2,jq2);
1449     qq23             = _mm_mul_pd(iq2,jq3);
1450     qq31             = _mm_mul_pd(iq3,jq1);
1451     qq32             = _mm_mul_pd(iq3,jq2);
1452     qq33             = _mm_mul_pd(iq3,jq3);
1453
1454     /* Avoid stupid compiler warnings */
1455     jnrA = jnrB = 0;
1456     j_coord_offsetA = 0;
1457     j_coord_offsetB = 0;
1458
1459     outeriter        = 0;
1460     inneriter        = 0;
1461
1462     /* Start outer loop over neighborlists */
1463     for(iidx=0; iidx<nri; iidx++)
1464     {
1465         /* Load shift vector for this list */
1466         i_shift_offset   = DIM*shiftidx[iidx];
1467
1468         /* Load limits for loop over neighbors */
1469         j_index_start    = jindex[iidx];
1470         j_index_end      = jindex[iidx+1];
1471
1472         /* Get outer coordinate index */
1473         inr              = iinr[iidx];
1474         i_coord_offset   = DIM*inr;
1475
1476         /* Load i particle coords and add shift vector */
1477         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1478                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1479
1480         fix0             = _mm_setzero_pd();
1481         fiy0             = _mm_setzero_pd();
1482         fiz0             = _mm_setzero_pd();
1483         fix1             = _mm_setzero_pd();
1484         fiy1             = _mm_setzero_pd();
1485         fiz1             = _mm_setzero_pd();
1486         fix2             = _mm_setzero_pd();
1487         fiy2             = _mm_setzero_pd();
1488         fiz2             = _mm_setzero_pd();
1489         fix3             = _mm_setzero_pd();
1490         fiy3             = _mm_setzero_pd();
1491         fiz3             = _mm_setzero_pd();
1492
1493         /* Start inner kernel loop */
1494         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1495         {
1496
1497             /* Get j neighbor index, and coordinate index */
1498             jnrA             = jjnr[jidx];
1499             jnrB             = jjnr[jidx+1];
1500             j_coord_offsetA  = DIM*jnrA;
1501             j_coord_offsetB  = DIM*jnrB;
1502
1503             /* load j atom coordinates */
1504             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1505                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1506                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1507
1508             /* Calculate displacement vector */
1509             dx00             = _mm_sub_pd(ix0,jx0);
1510             dy00             = _mm_sub_pd(iy0,jy0);
1511             dz00             = _mm_sub_pd(iz0,jz0);
1512             dx11             = _mm_sub_pd(ix1,jx1);
1513             dy11             = _mm_sub_pd(iy1,jy1);
1514             dz11             = _mm_sub_pd(iz1,jz1);
1515             dx12             = _mm_sub_pd(ix1,jx2);
1516             dy12             = _mm_sub_pd(iy1,jy2);
1517             dz12             = _mm_sub_pd(iz1,jz2);
1518             dx13             = _mm_sub_pd(ix1,jx3);
1519             dy13             = _mm_sub_pd(iy1,jy3);
1520             dz13             = _mm_sub_pd(iz1,jz3);
1521             dx21             = _mm_sub_pd(ix2,jx1);
1522             dy21             = _mm_sub_pd(iy2,jy1);
1523             dz21             = _mm_sub_pd(iz2,jz1);
1524             dx22             = _mm_sub_pd(ix2,jx2);
1525             dy22             = _mm_sub_pd(iy2,jy2);
1526             dz22             = _mm_sub_pd(iz2,jz2);
1527             dx23             = _mm_sub_pd(ix2,jx3);
1528             dy23             = _mm_sub_pd(iy2,jy3);
1529             dz23             = _mm_sub_pd(iz2,jz3);
1530             dx31             = _mm_sub_pd(ix3,jx1);
1531             dy31             = _mm_sub_pd(iy3,jy1);
1532             dz31             = _mm_sub_pd(iz3,jz1);
1533             dx32             = _mm_sub_pd(ix3,jx2);
1534             dy32             = _mm_sub_pd(iy3,jy2);
1535             dz32             = _mm_sub_pd(iz3,jz2);
1536             dx33             = _mm_sub_pd(ix3,jx3);
1537             dy33             = _mm_sub_pd(iy3,jy3);
1538             dz33             = _mm_sub_pd(iz3,jz3);
1539
1540             /* Calculate squared distance and things based on it */
1541             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1542             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1543             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1544             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1545             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1546             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1547             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1548             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1549             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1550             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1551
1552             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1553             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1554             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1555             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1556             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1557             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1558             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1559             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1560             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1561             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1562
1563             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1564             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1565             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1566             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1567             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1568             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1569             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1570             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1571             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1572             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1573
1574             fjx0             = _mm_setzero_pd();
1575             fjy0             = _mm_setzero_pd();
1576             fjz0             = _mm_setzero_pd();
1577             fjx1             = _mm_setzero_pd();
1578             fjy1             = _mm_setzero_pd();
1579             fjz1             = _mm_setzero_pd();
1580             fjx2             = _mm_setzero_pd();
1581             fjy2             = _mm_setzero_pd();
1582             fjz2             = _mm_setzero_pd();
1583             fjx3             = _mm_setzero_pd();
1584             fjy3             = _mm_setzero_pd();
1585             fjz3             = _mm_setzero_pd();
1586
1587             /**************************
1588              * CALCULATE INTERACTIONS *
1589              **************************/
1590
1591             r00              = _mm_mul_pd(rsq00,rinv00);
1592
1593             /* Analytical LJ-PME */
1594             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1595             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1596             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1597             exponent         = gmx_simd_exp_d(ewcljrsq);
1598             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1599             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1600             /* f6A = 6 * C6grid * (1 - poly) */
1601             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1602             /* f6B = C6grid * exponent * beta^6 */
1603             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1604             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1605             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1606
1607             fscal            = fvdw;
1608
1609             /* Update vectorial force */
1610             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1611             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1612             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1613
1614             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1615             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1616             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1617
1618             /**************************
1619              * CALCULATE INTERACTIONS *
1620              **************************/
1621
1622             r11              = _mm_mul_pd(rsq11,rinv11);
1623
1624             /* EWALD ELECTROSTATICS */
1625
1626             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1627             ewrt             = _mm_mul_pd(r11,ewtabscale);
1628             ewitab           = _mm_cvttpd_epi32(ewrt);
1629 #ifdef __XOP__
1630             eweps            = _mm_frcz_pd(ewrt);
1631 #else
1632             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1633 #endif
1634             twoeweps         = _mm_add_pd(eweps,eweps);
1635             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1636                                          &ewtabF,&ewtabFn);
1637             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1638             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1639
1640             fscal            = felec;
1641
1642             /* Update vectorial force */
1643             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1644             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1645             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1646
1647             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1648             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1649             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1650
1651             /**************************
1652              * CALCULATE INTERACTIONS *
1653              **************************/
1654
1655             r12              = _mm_mul_pd(rsq12,rinv12);
1656
1657             /* EWALD ELECTROSTATICS */
1658
1659             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1660             ewrt             = _mm_mul_pd(r12,ewtabscale);
1661             ewitab           = _mm_cvttpd_epi32(ewrt);
1662 #ifdef __XOP__
1663             eweps            = _mm_frcz_pd(ewrt);
1664 #else
1665             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1666 #endif
1667             twoeweps         = _mm_add_pd(eweps,eweps);
1668             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1669                                          &ewtabF,&ewtabFn);
1670             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1671             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1672
1673             fscal            = felec;
1674
1675             /* Update vectorial force */
1676             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1677             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1678             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1679
1680             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1681             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1682             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1683
1684             /**************************
1685              * CALCULATE INTERACTIONS *
1686              **************************/
1687
1688             r13              = _mm_mul_pd(rsq13,rinv13);
1689
1690             /* EWALD ELECTROSTATICS */
1691
1692             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1693             ewrt             = _mm_mul_pd(r13,ewtabscale);
1694             ewitab           = _mm_cvttpd_epi32(ewrt);
1695 #ifdef __XOP__
1696             eweps            = _mm_frcz_pd(ewrt);
1697 #else
1698             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1699 #endif
1700             twoeweps         = _mm_add_pd(eweps,eweps);
1701             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1702                                          &ewtabF,&ewtabFn);
1703             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1704             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1705
1706             fscal            = felec;
1707
1708             /* Update vectorial force */
1709             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1710             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1711             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1712
1713             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1714             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1715             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1716
1717             /**************************
1718              * CALCULATE INTERACTIONS *
1719              **************************/
1720
1721             r21              = _mm_mul_pd(rsq21,rinv21);
1722
1723             /* EWALD ELECTROSTATICS */
1724
1725             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1726             ewrt             = _mm_mul_pd(r21,ewtabscale);
1727             ewitab           = _mm_cvttpd_epi32(ewrt);
1728 #ifdef __XOP__
1729             eweps            = _mm_frcz_pd(ewrt);
1730 #else
1731             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1732 #endif
1733             twoeweps         = _mm_add_pd(eweps,eweps);
1734             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1735                                          &ewtabF,&ewtabFn);
1736             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1737             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1738
1739             fscal            = felec;
1740
1741             /* Update vectorial force */
1742             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1743             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1744             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1745
1746             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1747             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1748             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1749
1750             /**************************
1751              * CALCULATE INTERACTIONS *
1752              **************************/
1753
1754             r22              = _mm_mul_pd(rsq22,rinv22);
1755
1756             /* EWALD ELECTROSTATICS */
1757
1758             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1759             ewrt             = _mm_mul_pd(r22,ewtabscale);
1760             ewitab           = _mm_cvttpd_epi32(ewrt);
1761 #ifdef __XOP__
1762             eweps            = _mm_frcz_pd(ewrt);
1763 #else
1764             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1765 #endif
1766             twoeweps         = _mm_add_pd(eweps,eweps);
1767             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1768                                          &ewtabF,&ewtabFn);
1769             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1770             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1771
1772             fscal            = felec;
1773
1774             /* Update vectorial force */
1775             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1776             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1777             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1778
1779             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1780             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1781             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1782
1783             /**************************
1784              * CALCULATE INTERACTIONS *
1785              **************************/
1786
1787             r23              = _mm_mul_pd(rsq23,rinv23);
1788
1789             /* EWALD ELECTROSTATICS */
1790
1791             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1792             ewrt             = _mm_mul_pd(r23,ewtabscale);
1793             ewitab           = _mm_cvttpd_epi32(ewrt);
1794 #ifdef __XOP__
1795             eweps            = _mm_frcz_pd(ewrt);
1796 #else
1797             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1798 #endif
1799             twoeweps         = _mm_add_pd(eweps,eweps);
1800             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1801                                          &ewtabF,&ewtabFn);
1802             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1803             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1804
1805             fscal            = felec;
1806
1807             /* Update vectorial force */
1808             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1809             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1810             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1811
1812             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1813             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1814             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1815
1816             /**************************
1817              * CALCULATE INTERACTIONS *
1818              **************************/
1819
1820             r31              = _mm_mul_pd(rsq31,rinv31);
1821
1822             /* EWALD ELECTROSTATICS */
1823
1824             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1825             ewrt             = _mm_mul_pd(r31,ewtabscale);
1826             ewitab           = _mm_cvttpd_epi32(ewrt);
1827 #ifdef __XOP__
1828             eweps            = _mm_frcz_pd(ewrt);
1829 #else
1830             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1831 #endif
1832             twoeweps         = _mm_add_pd(eweps,eweps);
1833             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1834                                          &ewtabF,&ewtabFn);
1835             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1836             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1837
1838             fscal            = felec;
1839
1840             /* Update vectorial force */
1841             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1842             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1843             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1844
1845             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1846             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1847             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1848
1849             /**************************
1850              * CALCULATE INTERACTIONS *
1851              **************************/
1852
1853             r32              = _mm_mul_pd(rsq32,rinv32);
1854
1855             /* EWALD ELECTROSTATICS */
1856
1857             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1858             ewrt             = _mm_mul_pd(r32,ewtabscale);
1859             ewitab           = _mm_cvttpd_epi32(ewrt);
1860 #ifdef __XOP__
1861             eweps            = _mm_frcz_pd(ewrt);
1862 #else
1863             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1864 #endif
1865             twoeweps         = _mm_add_pd(eweps,eweps);
1866             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1867                                          &ewtabF,&ewtabFn);
1868             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1869             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1870
1871             fscal            = felec;
1872
1873             /* Update vectorial force */
1874             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1875             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1876             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1877
1878             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1879             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1880             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1881
1882             /**************************
1883              * CALCULATE INTERACTIONS *
1884              **************************/
1885
1886             r33              = _mm_mul_pd(rsq33,rinv33);
1887
1888             /* EWALD ELECTROSTATICS */
1889
1890             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1891             ewrt             = _mm_mul_pd(r33,ewtabscale);
1892             ewitab           = _mm_cvttpd_epi32(ewrt);
1893 #ifdef __XOP__
1894             eweps            = _mm_frcz_pd(ewrt);
1895 #else
1896             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1897 #endif
1898             twoeweps         = _mm_add_pd(eweps,eweps);
1899             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1900                                          &ewtabF,&ewtabFn);
1901             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1902             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1903
1904             fscal            = felec;
1905
1906             /* Update vectorial force */
1907             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1908             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1909             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1910
1911             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1912             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1913             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1914
1915             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);
1916
1917             /* Inner loop uses 401 flops */
1918         }
1919
1920         if(jidx<j_index_end)
1921         {
1922
1923             jnrA             = jjnr[jidx];
1924             j_coord_offsetA  = DIM*jnrA;
1925
1926             /* load j atom coordinates */
1927             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1928                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1929                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1930
1931             /* Calculate displacement vector */
1932             dx00             = _mm_sub_pd(ix0,jx0);
1933             dy00             = _mm_sub_pd(iy0,jy0);
1934             dz00             = _mm_sub_pd(iz0,jz0);
1935             dx11             = _mm_sub_pd(ix1,jx1);
1936             dy11             = _mm_sub_pd(iy1,jy1);
1937             dz11             = _mm_sub_pd(iz1,jz1);
1938             dx12             = _mm_sub_pd(ix1,jx2);
1939             dy12             = _mm_sub_pd(iy1,jy2);
1940             dz12             = _mm_sub_pd(iz1,jz2);
1941             dx13             = _mm_sub_pd(ix1,jx3);
1942             dy13             = _mm_sub_pd(iy1,jy3);
1943             dz13             = _mm_sub_pd(iz1,jz3);
1944             dx21             = _mm_sub_pd(ix2,jx1);
1945             dy21             = _mm_sub_pd(iy2,jy1);
1946             dz21             = _mm_sub_pd(iz2,jz1);
1947             dx22             = _mm_sub_pd(ix2,jx2);
1948             dy22             = _mm_sub_pd(iy2,jy2);
1949             dz22             = _mm_sub_pd(iz2,jz2);
1950             dx23             = _mm_sub_pd(ix2,jx3);
1951             dy23             = _mm_sub_pd(iy2,jy3);
1952             dz23             = _mm_sub_pd(iz2,jz3);
1953             dx31             = _mm_sub_pd(ix3,jx1);
1954             dy31             = _mm_sub_pd(iy3,jy1);
1955             dz31             = _mm_sub_pd(iz3,jz1);
1956             dx32             = _mm_sub_pd(ix3,jx2);
1957             dy32             = _mm_sub_pd(iy3,jy2);
1958             dz32             = _mm_sub_pd(iz3,jz2);
1959             dx33             = _mm_sub_pd(ix3,jx3);
1960             dy33             = _mm_sub_pd(iy3,jy3);
1961             dz33             = _mm_sub_pd(iz3,jz3);
1962
1963             /* Calculate squared distance and things based on it */
1964             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1965             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1966             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1967             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1968             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1969             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1970             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1971             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1972             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1973             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1974
1975             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1976             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1977             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1978             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1979             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1980             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1981             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1982             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1983             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1984             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1985
1986             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1987             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1988             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1989             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1990             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1991             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1992             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1993             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1994             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1995             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1996
1997             fjx0             = _mm_setzero_pd();
1998             fjy0             = _mm_setzero_pd();
1999             fjz0             = _mm_setzero_pd();
2000             fjx1             = _mm_setzero_pd();
2001             fjy1             = _mm_setzero_pd();
2002             fjz1             = _mm_setzero_pd();
2003             fjx2             = _mm_setzero_pd();
2004             fjy2             = _mm_setzero_pd();
2005             fjz2             = _mm_setzero_pd();
2006             fjx3             = _mm_setzero_pd();
2007             fjy3             = _mm_setzero_pd();
2008             fjz3             = _mm_setzero_pd();
2009
2010             /**************************
2011              * CALCULATE INTERACTIONS *
2012              **************************/
2013
2014             r00              = _mm_mul_pd(rsq00,rinv00);
2015
2016             /* Analytical LJ-PME */
2017             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2018             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
2019             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
2020             exponent         = gmx_simd_exp_d(ewcljrsq);
2021             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2022             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
2023             /* f6A = 6 * C6grid * (1 - poly) */
2024             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
2025             /* f6B = C6grid * exponent * beta^6 */
2026             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
2027             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2028             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
2029
2030             fscal            = fvdw;
2031
2032             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2033
2034             /* Update vectorial force */
2035             fix0             = _mm_macc_pd(dx00,fscal,fix0);
2036             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
2037             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
2038
2039             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
2040             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
2041             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
2042
2043             /**************************
2044              * CALCULATE INTERACTIONS *
2045              **************************/
2046
2047             r11              = _mm_mul_pd(rsq11,rinv11);
2048
2049             /* EWALD ELECTROSTATICS */
2050
2051             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2052             ewrt             = _mm_mul_pd(r11,ewtabscale);
2053             ewitab           = _mm_cvttpd_epi32(ewrt);
2054 #ifdef __XOP__
2055             eweps            = _mm_frcz_pd(ewrt);
2056 #else
2057             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2058 #endif
2059             twoeweps         = _mm_add_pd(eweps,eweps);
2060             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2061             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2062             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2063
2064             fscal            = felec;
2065
2066             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2067
2068             /* Update vectorial force */
2069             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2070             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2071             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2072
2073             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2074             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2075             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2076
2077             /**************************
2078              * CALCULATE INTERACTIONS *
2079              **************************/
2080
2081             r12              = _mm_mul_pd(rsq12,rinv12);
2082
2083             /* EWALD ELECTROSTATICS */
2084
2085             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2086             ewrt             = _mm_mul_pd(r12,ewtabscale);
2087             ewitab           = _mm_cvttpd_epi32(ewrt);
2088 #ifdef __XOP__
2089             eweps            = _mm_frcz_pd(ewrt);
2090 #else
2091             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2092 #endif
2093             twoeweps         = _mm_add_pd(eweps,eweps);
2094             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2095             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2096             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2097
2098             fscal            = felec;
2099
2100             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2101
2102             /* Update vectorial force */
2103             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2104             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2105             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2106
2107             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2108             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2109             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2110
2111             /**************************
2112              * CALCULATE INTERACTIONS *
2113              **************************/
2114
2115             r13              = _mm_mul_pd(rsq13,rinv13);
2116
2117             /* EWALD ELECTROSTATICS */
2118
2119             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2120             ewrt             = _mm_mul_pd(r13,ewtabscale);
2121             ewitab           = _mm_cvttpd_epi32(ewrt);
2122 #ifdef __XOP__
2123             eweps            = _mm_frcz_pd(ewrt);
2124 #else
2125             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2126 #endif
2127             twoeweps         = _mm_add_pd(eweps,eweps);
2128             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2129             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2130             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
2131
2132             fscal            = felec;
2133
2134             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2135
2136             /* Update vectorial force */
2137             fix1             = _mm_macc_pd(dx13,fscal,fix1);
2138             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
2139             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
2140
2141             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
2142             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
2143             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
2144
2145             /**************************
2146              * CALCULATE INTERACTIONS *
2147              **************************/
2148
2149             r21              = _mm_mul_pd(rsq21,rinv21);
2150
2151             /* EWALD ELECTROSTATICS */
2152
2153             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2154             ewrt             = _mm_mul_pd(r21,ewtabscale);
2155             ewitab           = _mm_cvttpd_epi32(ewrt);
2156 #ifdef __XOP__
2157             eweps            = _mm_frcz_pd(ewrt);
2158 #else
2159             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2160 #endif
2161             twoeweps         = _mm_add_pd(eweps,eweps);
2162             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2163             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2164             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2165
2166             fscal            = felec;
2167
2168             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2169
2170             /* Update vectorial force */
2171             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2172             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2173             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2174
2175             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2176             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2177             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2178
2179             /**************************
2180              * CALCULATE INTERACTIONS *
2181              **************************/
2182
2183             r22              = _mm_mul_pd(rsq22,rinv22);
2184
2185             /* EWALD ELECTROSTATICS */
2186
2187             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2188             ewrt             = _mm_mul_pd(r22,ewtabscale);
2189             ewitab           = _mm_cvttpd_epi32(ewrt);
2190 #ifdef __XOP__
2191             eweps            = _mm_frcz_pd(ewrt);
2192 #else
2193             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2194 #endif
2195             twoeweps         = _mm_add_pd(eweps,eweps);
2196             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2197             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2198             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2199
2200             fscal            = felec;
2201
2202             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2203
2204             /* Update vectorial force */
2205             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2206             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2207             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2208
2209             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2210             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2211             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2212
2213             /**************************
2214              * CALCULATE INTERACTIONS *
2215              **************************/
2216
2217             r23              = _mm_mul_pd(rsq23,rinv23);
2218
2219             /* EWALD ELECTROSTATICS */
2220
2221             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2222             ewrt             = _mm_mul_pd(r23,ewtabscale);
2223             ewitab           = _mm_cvttpd_epi32(ewrt);
2224 #ifdef __XOP__
2225             eweps            = _mm_frcz_pd(ewrt);
2226 #else
2227             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2228 #endif
2229             twoeweps         = _mm_add_pd(eweps,eweps);
2230             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2231             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2232             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
2233
2234             fscal            = felec;
2235
2236             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2237
2238             /* Update vectorial force */
2239             fix2             = _mm_macc_pd(dx23,fscal,fix2);
2240             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
2241             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
2242
2243             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
2244             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
2245             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
2246
2247             /**************************
2248              * CALCULATE INTERACTIONS *
2249              **************************/
2250
2251             r31              = _mm_mul_pd(rsq31,rinv31);
2252
2253             /* EWALD ELECTROSTATICS */
2254
2255             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2256             ewrt             = _mm_mul_pd(r31,ewtabscale);
2257             ewitab           = _mm_cvttpd_epi32(ewrt);
2258 #ifdef __XOP__
2259             eweps            = _mm_frcz_pd(ewrt);
2260 #else
2261             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2262 #endif
2263             twoeweps         = _mm_add_pd(eweps,eweps);
2264             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2265             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2266             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2267
2268             fscal            = felec;
2269
2270             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2271
2272             /* Update vectorial force */
2273             fix3             = _mm_macc_pd(dx31,fscal,fix3);
2274             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
2275             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
2276
2277             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
2278             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
2279             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
2280
2281             /**************************
2282              * CALCULATE INTERACTIONS *
2283              **************************/
2284
2285             r32              = _mm_mul_pd(rsq32,rinv32);
2286
2287             /* EWALD ELECTROSTATICS */
2288
2289             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2290             ewrt             = _mm_mul_pd(r32,ewtabscale);
2291             ewitab           = _mm_cvttpd_epi32(ewrt);
2292 #ifdef __XOP__
2293             eweps            = _mm_frcz_pd(ewrt);
2294 #else
2295             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2296 #endif
2297             twoeweps         = _mm_add_pd(eweps,eweps);
2298             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2299             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2300             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2301
2302             fscal            = felec;
2303
2304             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2305
2306             /* Update vectorial force */
2307             fix3             = _mm_macc_pd(dx32,fscal,fix3);
2308             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
2309             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
2310
2311             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
2312             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
2313             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
2314
2315             /**************************
2316              * CALCULATE INTERACTIONS *
2317              **************************/
2318
2319             r33              = _mm_mul_pd(rsq33,rinv33);
2320
2321             /* EWALD ELECTROSTATICS */
2322
2323             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2324             ewrt             = _mm_mul_pd(r33,ewtabscale);
2325             ewitab           = _mm_cvttpd_epi32(ewrt);
2326 #ifdef __XOP__
2327             eweps            = _mm_frcz_pd(ewrt);
2328 #else
2329             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2330 #endif
2331             twoeweps         = _mm_add_pd(eweps,eweps);
2332             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2333             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2334             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2335
2336             fscal            = felec;
2337
2338             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2339
2340             /* Update vectorial force */
2341             fix3             = _mm_macc_pd(dx33,fscal,fix3);
2342             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
2343             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
2344
2345             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
2346             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
2347             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
2348
2349             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2350
2351             /* Inner loop uses 401 flops */
2352         }
2353
2354         /* End of innermost loop */
2355
2356         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2357                                               f+i_coord_offset,fshift+i_shift_offset);
2358
2359         /* Increment number of inner iterations */
2360         inneriter                  += j_index_end - j_index_start;
2361
2362         /* Outer loop uses 24 flops */
2363     }
2364
2365     /* Increment number of outer iterations */
2366     outeriter        += nri;
2367
2368     /* Update outer/inner flops */
2369
2370     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*401);
2371 }