src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwCSTab_GeomW4W4_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4W4_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Water4
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwCSTab_GeomW4W4_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     int              vdwjidx1A,vdwjidx1B;
  90     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  91     int              vdwjidx2A,vdwjidx2B;
  92     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  93     int              vdwjidx3A,vdwjidx3B;
  94     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  95     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  97     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  98     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  99     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 100     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 101     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 102     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 103     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 104     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 105     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 106     real             *charge;
 107     int              nvdwtype;
 108     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 109     int              *vdwtype;
 110     real             *vdwparam;
 111     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 112     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 113     __m128i          vfitab;
 114     __m128i          ifour       = _mm_set1_epi32(4);
 115     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 116     real             *vftab;
 117     __m128i          ewitab;
 118     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 119     real             *ewtab;
 120     __m128d          dummy_mask,cutoff_mask;
 121     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 122     __m128d          one     = _mm_set1_pd(1.0);
 123     __m128d          two     = _mm_set1_pd(2.0);
 124     x                = xx[0];
 125     f                = ff[0];
 126
 127     nri              = nlist->nri;
 128     iinr             = nlist->iinr;
 129     jindex           = nlist->jindex;
 130     jjnr             = nlist->jjnr;
 131     shiftidx         = nlist->shift;
 132     gid              = nlist->gid;
 133     shiftvec         = fr->shift_vec[0];
 134     fshift           = fr->fshift[0];
 135     facel            = _mm_set1_pd(fr->ic->epsfac);
 136     charge           = mdatoms->chargeA;
 137     nvdwtype         = fr->ntype;
 138     vdwparam         = fr->nbfp;
 139     vdwtype          = mdatoms->typeA;
 140
 141     vftab            = kernel_data->table_vdw->data;
 142     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 143
 144     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 145     ewtab            = fr->ic->tabq_coul_FDV0;
 146     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 147     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 148
 149     /* Setup water-specific parameters */
 150     inr              = nlist->iinr[0];
 151     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 152     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 153     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 154     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 155
 156     jq1              = _mm_set1_pd(charge[inr+1]);
 157     jq2              = _mm_set1_pd(charge[inr+2]);
 158     jq3              = _mm_set1_pd(charge[inr+3]);
 159     vdwjidx0A        = 2*vdwtype[inr+0];
 160     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 161     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 162     qq11             = _mm_mul_pd(iq1,jq1);
 163     qq12             = _mm_mul_pd(iq1,jq2);
 164     qq13             = _mm_mul_pd(iq1,jq3);
 165     qq21             = _mm_mul_pd(iq2,jq1);
 166     qq22             = _mm_mul_pd(iq2,jq2);
 167     qq23             = _mm_mul_pd(iq2,jq3);
 168     qq31             = _mm_mul_pd(iq3,jq1);
 169     qq32             = _mm_mul_pd(iq3,jq2);
 170     qq33             = _mm_mul_pd(iq3,jq3);
 171
 172     /* Avoid stupid compiler warnings */
 173     jnrA = jnrB = 0;
 174     j_coord_offsetA = 0;
 175     j_coord_offsetB = 0;
 176
 177     outeriter        = 0;
 178     inneriter        = 0;
 179
 180     /* Start outer loop over neighborlists */
 181     for(iidx=0; iidx<nri; iidx++)
 182     {
 183         /* Load shift vector for this list */
 184         i_shift_offset   = DIM*shiftidx[iidx];
 185
 186         /* Load limits for loop over neighbors */
 187         j_index_start    = jindex[iidx];
 188         j_index_end      = jindex[iidx+1];
 189
 190         /* Get outer coordinate index */
 191         inr              = iinr[iidx];
 192         i_coord_offset   = DIM*inr;
 193
 194         /* Load i particle coords and add shift vector */
 195         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 196                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 197
 198         fix0             = _mm_setzero_pd();
 199         fiy0             = _mm_setzero_pd();
 200         fiz0             = _mm_setzero_pd();
 201         fix1             = _mm_setzero_pd();
 202         fiy1             = _mm_setzero_pd();
 203         fiz1             = _mm_setzero_pd();
 204         fix2             = _mm_setzero_pd();
 205         fiy2             = _mm_setzero_pd();
 206         fiz2             = _mm_setzero_pd();
 207         fix3             = _mm_setzero_pd();
 208         fiy3             = _mm_setzero_pd();
 209         fiz3             = _mm_setzero_pd();
 210
 211         /* Reset potential sums */
 212         velecsum         = _mm_setzero_pd();
 213         vvdwsum          = _mm_setzero_pd();
 214
 215         /* Start inner kernel loop */
 216         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 217         {
 218
 219             /* Get j neighbor index, and coordinate index */
 220             jnrA             = jjnr[jidx];
 221             jnrB             = jjnr[jidx+1];
 222             j_coord_offsetA  = DIM*jnrA;
 223             j_coord_offsetB  = DIM*jnrB;
 224
 225             /* load j atom coordinates */
 226             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 227                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 228                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 229
 230             /* Calculate displacement vector */
 231             dx00             = _mm_sub_pd(ix0,jx0);
 232             dy00             = _mm_sub_pd(iy0,jy0);
 233             dz00             = _mm_sub_pd(iz0,jz0);
 234             dx11             = _mm_sub_pd(ix1,jx1);
 235             dy11             = _mm_sub_pd(iy1,jy1);
 236             dz11             = _mm_sub_pd(iz1,jz1);
 237             dx12             = _mm_sub_pd(ix1,jx2);
 238             dy12             = _mm_sub_pd(iy1,jy2);
 239             dz12             = _mm_sub_pd(iz1,jz2);
 240             dx13             = _mm_sub_pd(ix1,jx3);
 241             dy13             = _mm_sub_pd(iy1,jy3);
 242             dz13             = _mm_sub_pd(iz1,jz3);
 243             dx21             = _mm_sub_pd(ix2,jx1);
 244             dy21             = _mm_sub_pd(iy2,jy1);
 245             dz21             = _mm_sub_pd(iz2,jz1);
 246             dx22             = _mm_sub_pd(ix2,jx2);
 247             dy22             = _mm_sub_pd(iy2,jy2);
 248             dz22             = _mm_sub_pd(iz2,jz2);
 249             dx23             = _mm_sub_pd(ix2,jx3);
 250             dy23             = _mm_sub_pd(iy2,jy3);
 251             dz23             = _mm_sub_pd(iz2,jz3);
 252             dx31             = _mm_sub_pd(ix3,jx1);
 253             dy31             = _mm_sub_pd(iy3,jy1);
 254             dz31             = _mm_sub_pd(iz3,jz1);
 255             dx32             = _mm_sub_pd(ix3,jx2);
 256             dy32             = _mm_sub_pd(iy3,jy2);
 257             dz32             = _mm_sub_pd(iz3,jz2);
 258             dx33             = _mm_sub_pd(ix3,jx3);
 259             dy33             = _mm_sub_pd(iy3,jy3);
 260             dz33             = _mm_sub_pd(iz3,jz3);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 264             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 265             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 266             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 267             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 268             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 269             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 270             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 271             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 272             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 273
 274             rinv00           = avx128fma_invsqrt_d(rsq00);
 275             rinv11           = avx128fma_invsqrt_d(rsq11);
 276             rinv12           = avx128fma_invsqrt_d(rsq12);
 277             rinv13           = avx128fma_invsqrt_d(rsq13);
 278             rinv21           = avx128fma_invsqrt_d(rsq21);
 279             rinv22           = avx128fma_invsqrt_d(rsq22);
 280             rinv23           = avx128fma_invsqrt_d(rsq23);
 281             rinv31           = avx128fma_invsqrt_d(rsq31);
 282             rinv32           = avx128fma_invsqrt_d(rsq32);
 283             rinv33           = avx128fma_invsqrt_d(rsq33);
 284
 285             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 286             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 287             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 288             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 289             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 290             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 291             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 292             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 293             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 294
 295             fjx0             = _mm_setzero_pd();
 296             fjy0             = _mm_setzero_pd();
 297             fjz0             = _mm_setzero_pd();
 298             fjx1             = _mm_setzero_pd();
 299             fjy1             = _mm_setzero_pd();
 300             fjz1             = _mm_setzero_pd();
 301             fjx2             = _mm_setzero_pd();
 302             fjy2             = _mm_setzero_pd();
 303             fjz2             = _mm_setzero_pd();
 304             fjx3             = _mm_setzero_pd();
 305             fjy3             = _mm_setzero_pd();
 306             fjz3             = _mm_setzero_pd();
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             r00              = _mm_mul_pd(rsq00,rinv00);
 313
 314             /* Calculate table index by multiplying r with table scale and truncate to integer */
 315             rt               = _mm_mul_pd(r00,vftabscale);
 316             vfitab           = _mm_cvttpd_epi32(rt);
 317 #ifdef __XOP__
 318             vfeps            = _mm_frcz_pd(rt);
 319 #else
 320             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 321 #endif
 322             twovfeps         = _mm_add_pd(vfeps,vfeps);
 323             vfitab           = _mm_slli_epi32(vfitab,3);
 324
 325             /* CUBIC SPLINE TABLE DISPERSION */
 326             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 327             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 328             GMX_MM_TRANSPOSE2_PD(Y,F);
 329             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 330             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 331             GMX_MM_TRANSPOSE2_PD(G,H);
 332             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 333             VV               = _mm_macc_pd(vfeps,Fp,Y);
 334             vvdw6            = _mm_mul_pd(c6_00,VV);
 335             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 336             fvdw6            = _mm_mul_pd(c6_00,FF);
 337
 338             /* CUBIC SPLINE TABLE REPULSION */
 339             vfitab           = _mm_add_epi32(vfitab,ifour);
 340             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 341             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 342             GMX_MM_TRANSPOSE2_PD(Y,F);
 343             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 344             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 345             GMX_MM_TRANSPOSE2_PD(G,H);
 346             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 347             VV               = _mm_macc_pd(vfeps,Fp,Y);
 348             vvdw12           = _mm_mul_pd(c12_00,VV);
 349             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 350             fvdw12           = _mm_mul_pd(c12_00,FF);
 351             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 352             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 356
 357             fscal            = fvdw;
 358
 359             /* Update vectorial force */
 360             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 361             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 362             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 363
 364             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 365             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 366             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             r11              = _mm_mul_pd(rsq11,rinv11);
 373
 374             /* EWALD ELECTROSTATICS */
 375
 376             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 377             ewrt             = _mm_mul_pd(r11,ewtabscale);
 378             ewitab           = _mm_cvttpd_epi32(ewrt);
 379 #ifdef __XOP__
 380             eweps            = _mm_frcz_pd(ewrt);
 381 #else
 382             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 383 #endif
 384             twoeweps         = _mm_add_pd(eweps,eweps);
 385             ewitab           = _mm_slli_epi32(ewitab,2);
 386             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 387             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 388             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 389             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 390             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 391             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 392             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 393             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 394             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 395             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 396
 397             /* Update potential sum for this i atom from the interaction with this j atom. */
 398             velecsum         = _mm_add_pd(velecsum,velec);
 399
 400             fscal            = felec;
 401
 402             /* Update vectorial force */
 403             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 404             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 405             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 406
 407             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 408             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 409             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 410
 411             /**************************
 412              * CALCULATE INTERACTIONS *
 413              **************************/
 414
 415             r12              = _mm_mul_pd(rsq12,rinv12);
 416
 417             /* EWALD ELECTROSTATICS */
 418
 419             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 420             ewrt             = _mm_mul_pd(r12,ewtabscale);
 421             ewitab           = _mm_cvttpd_epi32(ewrt);
 422 #ifdef __XOP__
 423             eweps            = _mm_frcz_pd(ewrt);
 424 #else
 425             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 426 #endif
 427             twoeweps         = _mm_add_pd(eweps,eweps);
 428             ewitab           = _mm_slli_epi32(ewitab,2);
 429             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 430             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 431             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 432             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 433             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 434             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 435             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 436             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 437             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 438             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 439
 440             /* Update potential sum for this i atom from the interaction with this j atom. */
 441             velecsum         = _mm_add_pd(velecsum,velec);
 442
 443             fscal            = felec;
 444
 445             /* Update vectorial force */
 446             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 447             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 448             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 449
 450             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 451             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 452             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             r13              = _mm_mul_pd(rsq13,rinv13);
 459
 460             /* EWALD ELECTROSTATICS */
 461
 462             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 463             ewrt             = _mm_mul_pd(r13,ewtabscale);
 464             ewitab           = _mm_cvttpd_epi32(ewrt);
 465 #ifdef __XOP__
 466             eweps            = _mm_frcz_pd(ewrt);
 467 #else
 468             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 469 #endif
 470             twoeweps         = _mm_add_pd(eweps,eweps);
 471             ewitab           = _mm_slli_epi32(ewitab,2);
 472             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 473             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 474             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 475             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 476             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 477             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 478             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 479             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 480             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
 481             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 482
 483             /* Update potential sum for this i atom from the interaction with this j atom. */
 484             velecsum         = _mm_add_pd(velecsum,velec);
 485
 486             fscal            = felec;
 487
 488             /* Update vectorial force */
 489             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 490             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 491             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 492
 493             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 494             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 495             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 496
 497             /**************************
 498              * CALCULATE INTERACTIONS *
 499              **************************/
 500
 501             r21              = _mm_mul_pd(rsq21,rinv21);
 502
 503             /* EWALD ELECTROSTATICS */
 504
 505             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 506             ewrt             = _mm_mul_pd(r21,ewtabscale);
 507             ewitab           = _mm_cvttpd_epi32(ewrt);
 508 #ifdef __XOP__
 509             eweps            = _mm_frcz_pd(ewrt);
 510 #else
 511             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 512 #endif
 513             twoeweps         = _mm_add_pd(eweps,eweps);
 514             ewitab           = _mm_slli_epi32(ewitab,2);
 515             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 516             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 517             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 518             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 519             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 520             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 521             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 522             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 523             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 524             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 525
 526             /* Update potential sum for this i atom from the interaction with this j atom. */
 527             velecsum         = _mm_add_pd(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531             /* Update vectorial force */
 532             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 533             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 534             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 535
 536             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 537             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 538             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             r22              = _mm_mul_pd(rsq22,rinv22);
 545
 546             /* EWALD ELECTROSTATICS */
 547
 548             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 549             ewrt             = _mm_mul_pd(r22,ewtabscale);
 550             ewitab           = _mm_cvttpd_epi32(ewrt);
 551 #ifdef __XOP__
 552             eweps            = _mm_frcz_pd(ewrt);
 553 #else
 554             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 555 #endif
 556             twoeweps         = _mm_add_pd(eweps,eweps);
 557             ewitab           = _mm_slli_epi32(ewitab,2);
 558             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 559             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 560             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 561             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 562             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 563             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 564             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 565             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 566             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 567             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velecsum         = _mm_add_pd(velecsum,velec);
 571
 572             fscal            = felec;
 573
 574             /* Update vectorial force */
 575             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 576             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 577             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 578
 579             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 580             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 581             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 582
 583             /**************************
 584              * CALCULATE INTERACTIONS *
 585              **************************/
 586
 587             r23              = _mm_mul_pd(rsq23,rinv23);
 588
 589             /* EWALD ELECTROSTATICS */
 590
 591             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 592             ewrt             = _mm_mul_pd(r23,ewtabscale);
 593             ewitab           = _mm_cvttpd_epi32(ewrt);
 594 #ifdef __XOP__
 595             eweps            = _mm_frcz_pd(ewrt);
 596 #else
 597             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 598 #endif
 599             twoeweps         = _mm_add_pd(eweps,eweps);
 600             ewitab           = _mm_slli_epi32(ewitab,2);
 601             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 602             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 603             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 604             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 605             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 606             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 607             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 608             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 609             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
 610             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
 611
 612             /* Update potential sum for this i atom from the interaction with this j atom. */
 613             velecsum         = _mm_add_pd(velecsum,velec);
 614
 615             fscal            = felec;
 616
 617             /* Update vectorial force */
 618             fix2             = _mm_macc_pd(dx23,fscal,fix2);
 619             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
 620             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
 621
 622             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
 623             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
 624             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             r31              = _mm_mul_pd(rsq31,rinv31);
 631
 632             /* EWALD ELECTROSTATICS */
 633
 634             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 635             ewrt             = _mm_mul_pd(r31,ewtabscale);
 636             ewitab           = _mm_cvttpd_epi32(ewrt);
 637 #ifdef __XOP__
 638             eweps            = _mm_frcz_pd(ewrt);
 639 #else
 640             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 641 #endif
 642             twoeweps         = _mm_add_pd(eweps,eweps);
 643             ewitab           = _mm_slli_epi32(ewitab,2);
 644             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 645             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 646             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 647             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 648             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 649             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 650             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 651             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 652             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
 653             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
 654
 655             /* Update potential sum for this i atom from the interaction with this j atom. */
 656             velecsum         = _mm_add_pd(velecsum,velec);
 657
 658             fscal            = felec;
 659
 660             /* Update vectorial force */
 661             fix3             = _mm_macc_pd(dx31,fscal,fix3);
 662             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
 663             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
 664
 665             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
 666             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
 667             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
 668
 669             /**************************
 670              * CALCULATE INTERACTIONS *
 671              **************************/
 672
 673             r32              = _mm_mul_pd(rsq32,rinv32);
 674
 675             /* EWALD ELECTROSTATICS */
 676
 677             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 678             ewrt             = _mm_mul_pd(r32,ewtabscale);
 679             ewitab           = _mm_cvttpd_epi32(ewrt);
 680 #ifdef __XOP__
 681             eweps            = _mm_frcz_pd(ewrt);
 682 #else
 683             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 684 #endif
 685             twoeweps         = _mm_add_pd(eweps,eweps);
 686             ewitab           = _mm_slli_epi32(ewitab,2);
 687             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 688             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 689             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 690             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 691             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 692             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 693             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 694             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 695             velec            = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
 696             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
 697
 698             /* Update potential sum for this i atom from the interaction with this j atom. */
 699             velecsum         = _mm_add_pd(velecsum,velec);
 700
 701             fscal            = felec;
 702
 703             /* Update vectorial force */
 704             fix3             = _mm_macc_pd(dx32,fscal,fix3);
 705             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
 706             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
 707
 708             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
 709             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
 710             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
 711
 712             /**************************
 713              * CALCULATE INTERACTIONS *
 714              **************************/
 715
 716             r33              = _mm_mul_pd(rsq33,rinv33);
 717
 718             /* EWALD ELECTROSTATICS */
 719
 720             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 721             ewrt             = _mm_mul_pd(r33,ewtabscale);
 722             ewitab           = _mm_cvttpd_epi32(ewrt);
 723 #ifdef __XOP__
 724             eweps            = _mm_frcz_pd(ewrt);
 725 #else
 726             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 727 #endif
 728             twoeweps         = _mm_add_pd(eweps,eweps);
 729             ewitab           = _mm_slli_epi32(ewitab,2);
 730             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 731             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 732             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 733             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 734             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 735             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 736             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 737             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 738             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
 739             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
 740
 741             /* Update potential sum for this i atom from the interaction with this j atom. */
 742             velecsum         = _mm_add_pd(velecsum,velec);
 743
 744             fscal            = felec;
 745
 746             /* Update vectorial force */
 747             fix3             = _mm_macc_pd(dx33,fscal,fix3);
 748             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
 749             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
 750
 751             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
 752             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
 753             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
 754
 755             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);
 756
 757             /* Inner loop uses 458 flops */
 758         }
 759
 760         if(jidx<j_index_end)
 761         {
 762
 763             jnrA             = jjnr[jidx];
 764             j_coord_offsetA  = DIM*jnrA;
 765
 766             /* load j atom coordinates */
 767             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 768                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 769                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 770
 771             /* Calculate displacement vector */
 772             dx00             = _mm_sub_pd(ix0,jx0);
 773             dy00             = _mm_sub_pd(iy0,jy0);
 774             dz00             = _mm_sub_pd(iz0,jz0);
 775             dx11             = _mm_sub_pd(ix1,jx1);
 776             dy11             = _mm_sub_pd(iy1,jy1);
 777             dz11             = _mm_sub_pd(iz1,jz1);
 778             dx12             = _mm_sub_pd(ix1,jx2);
 779             dy12             = _mm_sub_pd(iy1,jy2);
 780             dz12             = _mm_sub_pd(iz1,jz2);
 781             dx13             = _mm_sub_pd(ix1,jx3);
 782             dy13             = _mm_sub_pd(iy1,jy3);
 783             dz13             = _mm_sub_pd(iz1,jz3);
 784             dx21             = _mm_sub_pd(ix2,jx1);
 785             dy21             = _mm_sub_pd(iy2,jy1);
 786             dz21             = _mm_sub_pd(iz2,jz1);
 787             dx22             = _mm_sub_pd(ix2,jx2);
 788             dy22             = _mm_sub_pd(iy2,jy2);
 789             dz22             = _mm_sub_pd(iz2,jz2);
 790             dx23             = _mm_sub_pd(ix2,jx3);
 791             dy23             = _mm_sub_pd(iy2,jy3);
 792             dz23             = _mm_sub_pd(iz2,jz3);
 793             dx31             = _mm_sub_pd(ix3,jx1);
 794             dy31             = _mm_sub_pd(iy3,jy1);
 795             dz31             = _mm_sub_pd(iz3,jz1);
 796             dx32             = _mm_sub_pd(ix3,jx2);
 797             dy32             = _mm_sub_pd(iy3,jy2);
 798             dz32             = _mm_sub_pd(iz3,jz2);
 799             dx33             = _mm_sub_pd(ix3,jx3);
 800             dy33             = _mm_sub_pd(iy3,jy3);
 801             dz33             = _mm_sub_pd(iz3,jz3);
 802
 803             /* Calculate squared distance and things based on it */
 804             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 805             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 806             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 807             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 808             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 809             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 810             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 811             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 812             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 813             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 814
 815             rinv00           = avx128fma_invsqrt_d(rsq00);
 816             rinv11           = avx128fma_invsqrt_d(rsq11);
 817             rinv12           = avx128fma_invsqrt_d(rsq12);
 818             rinv13           = avx128fma_invsqrt_d(rsq13);
 819             rinv21           = avx128fma_invsqrt_d(rsq21);
 820             rinv22           = avx128fma_invsqrt_d(rsq22);
 821             rinv23           = avx128fma_invsqrt_d(rsq23);
 822             rinv31           = avx128fma_invsqrt_d(rsq31);
 823             rinv32           = avx128fma_invsqrt_d(rsq32);
 824             rinv33           = avx128fma_invsqrt_d(rsq33);
 825
 826             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 827             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 828             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 829             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 830             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 831             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 832             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 833             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 834             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 835
 836             fjx0             = _mm_setzero_pd();
 837             fjy0             = _mm_setzero_pd();
 838             fjz0             = _mm_setzero_pd();
 839             fjx1             = _mm_setzero_pd();
 840             fjy1             = _mm_setzero_pd();
 841             fjz1             = _mm_setzero_pd();
 842             fjx2             = _mm_setzero_pd();
 843             fjy2             = _mm_setzero_pd();
 844             fjz2             = _mm_setzero_pd();
 845             fjx3             = _mm_setzero_pd();
 846             fjy3             = _mm_setzero_pd();
 847             fjz3             = _mm_setzero_pd();
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             r00              = _mm_mul_pd(rsq00,rinv00);
 854
 855             /* Calculate table index by multiplying r with table scale and truncate to integer */
 856             rt               = _mm_mul_pd(r00,vftabscale);
 857             vfitab           = _mm_cvttpd_epi32(rt);
 858 #ifdef __XOP__
 859             vfeps            = _mm_frcz_pd(rt);
 860 #else
 861             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 862 #endif
 863             twovfeps         = _mm_add_pd(vfeps,vfeps);
 864             vfitab           = _mm_slli_epi32(vfitab,3);
 865
 866             /* CUBIC SPLINE TABLE DISPERSION */
 867             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 868             F                = _mm_setzero_pd();
 869             GMX_MM_TRANSPOSE2_PD(Y,F);
 870             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 871             H                = _mm_setzero_pd();
 872             GMX_MM_TRANSPOSE2_PD(G,H);
 873             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 874             VV               = _mm_macc_pd(vfeps,Fp,Y);
 875             vvdw6            = _mm_mul_pd(c6_00,VV);
 876             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 877             fvdw6            = _mm_mul_pd(c6_00,FF);
 878
 879             /* CUBIC SPLINE TABLE REPULSION */
 880             vfitab           = _mm_add_epi32(vfitab,ifour);
 881             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 882             F                = _mm_setzero_pd();
 883             GMX_MM_TRANSPOSE2_PD(Y,F);
 884             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 885             H                = _mm_setzero_pd();
 886             GMX_MM_TRANSPOSE2_PD(G,H);
 887             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 888             VV               = _mm_macc_pd(vfeps,Fp,Y);
 889             vvdw12           = _mm_mul_pd(c12_00,VV);
 890             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 891             fvdw12           = _mm_mul_pd(c12_00,FF);
 892             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 893             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 894
 895             /* Update potential sum for this i atom from the interaction with this j atom. */
 896             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 897             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 898
 899             fscal            = fvdw;
 900
 901             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 902
 903             /* Update vectorial force */
 904             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 905             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 906             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 907
 908             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 909             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 910             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 911
 912             /**************************
 913              * CALCULATE INTERACTIONS *
 914              **************************/
 915
 916             r11              = _mm_mul_pd(rsq11,rinv11);
 917
 918             /* EWALD ELECTROSTATICS */
 919
 920             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 921             ewrt             = _mm_mul_pd(r11,ewtabscale);
 922             ewitab           = _mm_cvttpd_epi32(ewrt);
 923 #ifdef __XOP__
 924             eweps            = _mm_frcz_pd(ewrt);
 925 #else
 926             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 927 #endif
 928             twoeweps         = _mm_add_pd(eweps,eweps);
 929             ewitab           = _mm_slli_epi32(ewitab,2);
 930             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 931             ewtabD           = _mm_setzero_pd();
 932             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 933             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 934             ewtabFn          = _mm_setzero_pd();
 935             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 936             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 937             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 938             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 939             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 940
 941             /* Update potential sum for this i atom from the interaction with this j atom. */
 942             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 943             velecsum         = _mm_add_pd(velecsum,velec);
 944
 945             fscal            = felec;
 946
 947             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 948
 949             /* Update vectorial force */
 950             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 951             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 952             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 953
 954             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 955             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 956             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             r12              = _mm_mul_pd(rsq12,rinv12);
 963
 964             /* EWALD ELECTROSTATICS */
 965
 966             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 967             ewrt             = _mm_mul_pd(r12,ewtabscale);
 968             ewitab           = _mm_cvttpd_epi32(ewrt);
 969 #ifdef __XOP__
 970             eweps            = _mm_frcz_pd(ewrt);
 971 #else
 972             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 973 #endif
 974             twoeweps         = _mm_add_pd(eweps,eweps);
 975             ewitab           = _mm_slli_epi32(ewitab,2);
 976             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 977             ewtabD           = _mm_setzero_pd();
 978             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 979             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 980             ewtabFn          = _mm_setzero_pd();
 981             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 982             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 983             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 984             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 985             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 986
 987             /* Update potential sum for this i atom from the interaction with this j atom. */
 988             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 989             velecsum         = _mm_add_pd(velecsum,velec);
 990
 991             fscal            = felec;
 992
 993             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 994
 995             /* Update vectorial force */
 996             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 997             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 998             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 999
1000             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1001             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1002             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             r13              = _mm_mul_pd(rsq13,rinv13);
1009
1010             /* EWALD ELECTROSTATICS */
1011
1012             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1013             ewrt             = _mm_mul_pd(r13,ewtabscale);
1014             ewitab           = _mm_cvttpd_epi32(ewrt);
1015 #ifdef __XOP__
1016             eweps            = _mm_frcz_pd(ewrt);
1017 #else
1018             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1019 #endif
1020             twoeweps         = _mm_add_pd(eweps,eweps);
1021             ewitab           = _mm_slli_epi32(ewitab,2);
1022             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1023             ewtabD           = _mm_setzero_pd();
1024             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1025             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1026             ewtabFn          = _mm_setzero_pd();
1027             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1028             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1029             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1030             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
1031             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1032
1033             /* Update potential sum for this i atom from the interaction with this j atom. */
1034             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1035             velecsum         = _mm_add_pd(velecsum,velec);
1036
1037             fscal            = felec;
1038
1039             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1040
1041             /* Update vectorial force */
1042             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1043             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1044             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1045
1046             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1047             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1048             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             r21              = _mm_mul_pd(rsq21,rinv21);
1055
1056             /* EWALD ELECTROSTATICS */
1057
1058             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1059             ewrt             = _mm_mul_pd(r21,ewtabscale);
1060             ewitab           = _mm_cvttpd_epi32(ewrt);
1061 #ifdef __XOP__
1062             eweps            = _mm_frcz_pd(ewrt);
1063 #else
1064             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1065 #endif
1066             twoeweps         = _mm_add_pd(eweps,eweps);
1067             ewitab           = _mm_slli_epi32(ewitab,2);
1068             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1069             ewtabD           = _mm_setzero_pd();
1070             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1071             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1072             ewtabFn          = _mm_setzero_pd();
1073             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1074             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1075             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1076             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1077             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1078
1079             /* Update potential sum for this i atom from the interaction with this j atom. */
1080             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1081             velecsum         = _mm_add_pd(velecsum,velec);
1082
1083             fscal            = felec;
1084
1085             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1086
1087             /* Update vectorial force */
1088             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1089             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1090             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1091
1092             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1093             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1094             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             r22              = _mm_mul_pd(rsq22,rinv22);
1101
1102             /* EWALD ELECTROSTATICS */
1103
1104             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1105             ewrt             = _mm_mul_pd(r22,ewtabscale);
1106             ewitab           = _mm_cvttpd_epi32(ewrt);
1107 #ifdef __XOP__
1108             eweps            = _mm_frcz_pd(ewrt);
1109 #else
1110             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1111 #endif
1112             twoeweps         = _mm_add_pd(eweps,eweps);
1113             ewitab           = _mm_slli_epi32(ewitab,2);
1114             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1115             ewtabD           = _mm_setzero_pd();
1116             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1117             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1118             ewtabFn          = _mm_setzero_pd();
1119             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1120             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1121             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1122             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1123             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1124
1125             /* Update potential sum for this i atom from the interaction with this j atom. */
1126             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1127             velecsum         = _mm_add_pd(velecsum,velec);
1128
1129             fscal            = felec;
1130
1131             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1132
1133             /* Update vectorial force */
1134             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1135             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1136             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1137
1138             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1139             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1140             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1141
1142             /**************************
1143              * CALCULATE INTERACTIONS *
1144              **************************/
1145
1146             r23              = _mm_mul_pd(rsq23,rinv23);
1147
1148             /* EWALD ELECTROSTATICS */
1149
1150             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1151             ewrt             = _mm_mul_pd(r23,ewtabscale);
1152             ewitab           = _mm_cvttpd_epi32(ewrt);
1153 #ifdef __XOP__
1154             eweps            = _mm_frcz_pd(ewrt);
1155 #else
1156             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1157 #endif
1158             twoeweps         = _mm_add_pd(eweps,eweps);
1159             ewitab           = _mm_slli_epi32(ewitab,2);
1160             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1161             ewtabD           = _mm_setzero_pd();
1162             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1163             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1164             ewtabFn          = _mm_setzero_pd();
1165             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1166             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1167             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1168             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
1169             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1170
1171             /* Update potential sum for this i atom from the interaction with this j atom. */
1172             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1173             velecsum         = _mm_add_pd(velecsum,velec);
1174
1175             fscal            = felec;
1176
1177             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1178
1179             /* Update vectorial force */
1180             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1181             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1182             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1183
1184             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1185             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1186             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             r31              = _mm_mul_pd(rsq31,rinv31);
1193
1194             /* EWALD ELECTROSTATICS */
1195
1196             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1197             ewrt             = _mm_mul_pd(r31,ewtabscale);
1198             ewitab           = _mm_cvttpd_epi32(ewrt);
1199 #ifdef __XOP__
1200             eweps            = _mm_frcz_pd(ewrt);
1201 #else
1202             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1203 #endif
1204             twoeweps         = _mm_add_pd(eweps,eweps);
1205             ewitab           = _mm_slli_epi32(ewitab,2);
1206             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1207             ewtabD           = _mm_setzero_pd();
1208             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1209             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1210             ewtabFn          = _mm_setzero_pd();
1211             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1212             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1213             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1214             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
1215             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1216
1217             /* Update potential sum for this i atom from the interaction with this j atom. */
1218             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1219             velecsum         = _mm_add_pd(velecsum,velec);
1220
1221             fscal            = felec;
1222
1223             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1224
1225             /* Update vectorial force */
1226             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1227             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1228             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1229
1230             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1231             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1232             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1233
1234             /**************************
1235              * CALCULATE INTERACTIONS *
1236              **************************/
1237
1238             r32              = _mm_mul_pd(rsq32,rinv32);
1239
1240             /* EWALD ELECTROSTATICS */
1241
1242             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1243             ewrt             = _mm_mul_pd(r32,ewtabscale);
1244             ewitab           = _mm_cvttpd_epi32(ewrt);
1245 #ifdef __XOP__
1246             eweps            = _mm_frcz_pd(ewrt);
1247 #else
1248             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1249 #endif
1250             twoeweps         = _mm_add_pd(eweps,eweps);
1251             ewitab           = _mm_slli_epi32(ewitab,2);
1252             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1253             ewtabD           = _mm_setzero_pd();
1254             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1255             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1256             ewtabFn          = _mm_setzero_pd();
1257             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1258             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1259             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1260             velec            = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
1261             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1262
1263             /* Update potential sum for this i atom from the interaction with this j atom. */
1264             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1265             velecsum         = _mm_add_pd(velecsum,velec);
1266
1267             fscal            = felec;
1268
1269             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1270
1271             /* Update vectorial force */
1272             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1273             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1274             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1275
1276             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1277             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1278             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1279
1280             /**************************
1281              * CALCULATE INTERACTIONS *
1282              **************************/
1283
1284             r33              = _mm_mul_pd(rsq33,rinv33);
1285
1286             /* EWALD ELECTROSTATICS */
1287
1288             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1289             ewrt             = _mm_mul_pd(r33,ewtabscale);
1290             ewitab           = _mm_cvttpd_epi32(ewrt);
1291 #ifdef __XOP__
1292             eweps            = _mm_frcz_pd(ewrt);
1293 #else
1294             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1295 #endif
1296             twoeweps         = _mm_add_pd(eweps,eweps);
1297             ewitab           = _mm_slli_epi32(ewitab,2);
1298             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1299             ewtabD           = _mm_setzero_pd();
1300             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1301             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1302             ewtabFn          = _mm_setzero_pd();
1303             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1304             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1305             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1306             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
1307             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1308
1309             /* Update potential sum for this i atom from the interaction with this j atom. */
1310             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1311             velecsum         = _mm_add_pd(velecsum,velec);
1312
1313             fscal            = felec;
1314
1315             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1316
1317             /* Update vectorial force */
1318             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1319             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1320             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1321
1322             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1323             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1324             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1325
1326             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1327
1328             /* Inner loop uses 458 flops */
1329         }
1330
1331         /* End of innermost loop */
1332
1333         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1334                                               f+i_coord_offset,fshift+i_shift_offset);
1335
1336         ggid                        = gid[iidx];
1337         /* Update potential energies */
1338         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1339         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1340
1341         /* Increment number of inner iterations */
1342         inneriter                  += j_index_end - j_index_start;
1343
1344         /* Outer loop uses 26 flops */
1345     }
1346
1347     /* Increment number of outer iterations */
1348     outeriter        += nri;
1349
1350     /* Update outer/inner flops */
1351
1352     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*458);
1353 }
1354 /*
1355  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4W4_F_avx_128_fma_double
1356  * Electrostatics interaction: Ewald
1357  * VdW interaction:            CubicSplineTable
1358  * Geometry:                   Water4-Water4
1359  * Calculate force/pot:        Force
1360  */
1361 void
1362 nb_kernel_ElecEw_VdwCSTab_GeomW4W4_F_avx_128_fma_double
1363                     (t_nblist                    * gmx_restrict       nlist,
1364                      rvec                        * gmx_restrict          xx,
1365                      rvec                        * gmx_restrict          ff,
1366                      struct t_forcerec           * gmx_restrict          fr,
1367                      t_mdatoms                   * gmx_restrict     mdatoms,
1368                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1369                      t_nrnb                      * gmx_restrict        nrnb)
1370 {
1371     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1372      * just 0 for non-waters.
1373      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1374      * jnr indices corresponding to data put in the four positions in the SIMD register.
1375      */
1376     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1377     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1378     int              jnrA,jnrB;
1379     int              j_coord_offsetA,j_coord_offsetB;
1380     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1381     real             rcutoff_scalar;
1382     real             *shiftvec,*fshift,*x,*f;
1383     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1384     int              vdwioffset0;
1385     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1386     int              vdwioffset1;
1387     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1388     int              vdwioffset2;
1389     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1390     int              vdwioffset3;
1391     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1392     int              vdwjidx0A,vdwjidx0B;
1393     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1394     int              vdwjidx1A,vdwjidx1B;
1395     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1396     int              vdwjidx2A,vdwjidx2B;
1397     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1398     int              vdwjidx3A,vdwjidx3B;
1399     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1400     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1401     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1402     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1403     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1404     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1405     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1406     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1407     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1408     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1409     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1410     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1411     real             *charge;
1412     int              nvdwtype;
1413     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1414     int              *vdwtype;
1415     real             *vdwparam;
1416     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1417     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1418     __m128i          vfitab;
1419     __m128i          ifour       = _mm_set1_epi32(4);
1420     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
1421     real             *vftab;
1422     __m128i          ewitab;
1423     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1424     real             *ewtab;
1425     __m128d          dummy_mask,cutoff_mask;
1426     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1427     __m128d          one     = _mm_set1_pd(1.0);
1428     __m128d          two     = _mm_set1_pd(2.0);
1429     x                = xx[0];
1430     f                = ff[0];
1431
1432     nri              = nlist->nri;
1433     iinr             = nlist->iinr;
1434     jindex           = nlist->jindex;
1435     jjnr             = nlist->jjnr;
1436     shiftidx         = nlist->shift;
1437     gid              = nlist->gid;
1438     shiftvec         = fr->shift_vec[0];
1439     fshift           = fr->fshift[0];
1440     facel            = _mm_set1_pd(fr->ic->epsfac);
1441     charge           = mdatoms->chargeA;
1442     nvdwtype         = fr->ntype;
1443     vdwparam         = fr->nbfp;
1444     vdwtype          = mdatoms->typeA;
1445
1446     vftab            = kernel_data->table_vdw->data;
1447     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
1448
1449     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1450     ewtab            = fr->ic->tabq_coul_F;
1451     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1452     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1453
1454     /* Setup water-specific parameters */
1455     inr              = nlist->iinr[0];
1456     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1457     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1458     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1459     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1460
1461     jq1              = _mm_set1_pd(charge[inr+1]);
1462     jq2              = _mm_set1_pd(charge[inr+2]);
1463     jq3              = _mm_set1_pd(charge[inr+3]);
1464     vdwjidx0A        = 2*vdwtype[inr+0];
1465     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1466     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1467     qq11             = _mm_mul_pd(iq1,jq1);
1468     qq12             = _mm_mul_pd(iq1,jq2);
1469     qq13             = _mm_mul_pd(iq1,jq3);
1470     qq21             = _mm_mul_pd(iq2,jq1);
1471     qq22             = _mm_mul_pd(iq2,jq2);
1472     qq23             = _mm_mul_pd(iq2,jq3);
1473     qq31             = _mm_mul_pd(iq3,jq1);
1474     qq32             = _mm_mul_pd(iq3,jq2);
1475     qq33             = _mm_mul_pd(iq3,jq3);
1476
1477     /* Avoid stupid compiler warnings */
1478     jnrA = jnrB = 0;
1479     j_coord_offsetA = 0;
1480     j_coord_offsetB = 0;
1481
1482     outeriter        = 0;
1483     inneriter        = 0;
1484
1485     /* Start outer loop over neighborlists */
1486     for(iidx=0; iidx<nri; iidx++)
1487     {
1488         /* Load shift vector for this list */
1489         i_shift_offset   = DIM*shiftidx[iidx];
1490
1491         /* Load limits for loop over neighbors */
1492         j_index_start    = jindex[iidx];
1493         j_index_end      = jindex[iidx+1];
1494
1495         /* Get outer coordinate index */
1496         inr              = iinr[iidx];
1497         i_coord_offset   = DIM*inr;
1498
1499         /* Load i particle coords and add shift vector */
1500         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1501                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1502
1503         fix0             = _mm_setzero_pd();
1504         fiy0             = _mm_setzero_pd();
1505         fiz0             = _mm_setzero_pd();
1506         fix1             = _mm_setzero_pd();
1507         fiy1             = _mm_setzero_pd();
1508         fiz1             = _mm_setzero_pd();
1509         fix2             = _mm_setzero_pd();
1510         fiy2             = _mm_setzero_pd();
1511         fiz2             = _mm_setzero_pd();
1512         fix3             = _mm_setzero_pd();
1513         fiy3             = _mm_setzero_pd();
1514         fiz3             = _mm_setzero_pd();
1515
1516         /* Start inner kernel loop */
1517         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1518         {
1519
1520             /* Get j neighbor index, and coordinate index */
1521             jnrA             = jjnr[jidx];
1522             jnrB             = jjnr[jidx+1];
1523             j_coord_offsetA  = DIM*jnrA;
1524             j_coord_offsetB  = DIM*jnrB;
1525
1526             /* load j atom coordinates */
1527             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1528                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1529                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1530
1531             /* Calculate displacement vector */
1532             dx00             = _mm_sub_pd(ix0,jx0);
1533             dy00             = _mm_sub_pd(iy0,jy0);
1534             dz00             = _mm_sub_pd(iz0,jz0);
1535             dx11             = _mm_sub_pd(ix1,jx1);
1536             dy11             = _mm_sub_pd(iy1,jy1);
1537             dz11             = _mm_sub_pd(iz1,jz1);
1538             dx12             = _mm_sub_pd(ix1,jx2);
1539             dy12             = _mm_sub_pd(iy1,jy2);
1540             dz12             = _mm_sub_pd(iz1,jz2);
1541             dx13             = _mm_sub_pd(ix1,jx3);
1542             dy13             = _mm_sub_pd(iy1,jy3);
1543             dz13             = _mm_sub_pd(iz1,jz3);
1544             dx21             = _mm_sub_pd(ix2,jx1);
1545             dy21             = _mm_sub_pd(iy2,jy1);
1546             dz21             = _mm_sub_pd(iz2,jz1);
1547             dx22             = _mm_sub_pd(ix2,jx2);
1548             dy22             = _mm_sub_pd(iy2,jy2);
1549             dz22             = _mm_sub_pd(iz2,jz2);
1550             dx23             = _mm_sub_pd(ix2,jx3);
1551             dy23             = _mm_sub_pd(iy2,jy3);
1552             dz23             = _mm_sub_pd(iz2,jz3);
1553             dx31             = _mm_sub_pd(ix3,jx1);
1554             dy31             = _mm_sub_pd(iy3,jy1);
1555             dz31             = _mm_sub_pd(iz3,jz1);
1556             dx32             = _mm_sub_pd(ix3,jx2);
1557             dy32             = _mm_sub_pd(iy3,jy2);
1558             dz32             = _mm_sub_pd(iz3,jz2);
1559             dx33             = _mm_sub_pd(ix3,jx3);
1560             dy33             = _mm_sub_pd(iy3,jy3);
1561             dz33             = _mm_sub_pd(iz3,jz3);
1562
1563             /* Calculate squared distance and things based on it */
1564             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1565             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1566             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1567             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1568             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1569             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1570             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1571             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1572             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1573             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1574
1575             rinv00           = avx128fma_invsqrt_d(rsq00);
1576             rinv11           = avx128fma_invsqrt_d(rsq11);
1577             rinv12           = avx128fma_invsqrt_d(rsq12);
1578             rinv13           = avx128fma_invsqrt_d(rsq13);
1579             rinv21           = avx128fma_invsqrt_d(rsq21);
1580             rinv22           = avx128fma_invsqrt_d(rsq22);
1581             rinv23           = avx128fma_invsqrt_d(rsq23);
1582             rinv31           = avx128fma_invsqrt_d(rsq31);
1583             rinv32           = avx128fma_invsqrt_d(rsq32);
1584             rinv33           = avx128fma_invsqrt_d(rsq33);
1585
1586             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1587             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1588             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1589             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1590             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1591             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1592             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1593             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1594             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1595
1596             fjx0             = _mm_setzero_pd();
1597             fjy0             = _mm_setzero_pd();
1598             fjz0             = _mm_setzero_pd();
1599             fjx1             = _mm_setzero_pd();
1600             fjy1             = _mm_setzero_pd();
1601             fjz1             = _mm_setzero_pd();
1602             fjx2             = _mm_setzero_pd();
1603             fjy2             = _mm_setzero_pd();
1604             fjz2             = _mm_setzero_pd();
1605             fjx3             = _mm_setzero_pd();
1606             fjy3             = _mm_setzero_pd();
1607             fjz3             = _mm_setzero_pd();
1608
1609             /**************************
1610              * CALCULATE INTERACTIONS *
1611              **************************/
1612
1613             r00              = _mm_mul_pd(rsq00,rinv00);
1614
1615             /* Calculate table index by multiplying r with table scale and truncate to integer */
1616             rt               = _mm_mul_pd(r00,vftabscale);
1617             vfitab           = _mm_cvttpd_epi32(rt);
1618 #ifdef __XOP__
1619             vfeps            = _mm_frcz_pd(rt);
1620 #else
1621             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1622 #endif
1623             twovfeps         = _mm_add_pd(vfeps,vfeps);
1624             vfitab           = _mm_slli_epi32(vfitab,3);
1625
1626             /* CUBIC SPLINE TABLE DISPERSION */
1627             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1628             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1629             GMX_MM_TRANSPOSE2_PD(Y,F);
1630             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1631             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1632             GMX_MM_TRANSPOSE2_PD(G,H);
1633             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1634             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1635             fvdw6            = _mm_mul_pd(c6_00,FF);
1636
1637             /* CUBIC SPLINE TABLE REPULSION */
1638             vfitab           = _mm_add_epi32(vfitab,ifour);
1639             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1640             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1641             GMX_MM_TRANSPOSE2_PD(Y,F);
1642             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1643             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1644             GMX_MM_TRANSPOSE2_PD(G,H);
1645             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1646             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1647             fvdw12           = _mm_mul_pd(c12_00,FF);
1648             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1649
1650             fscal            = fvdw;
1651
1652             /* Update vectorial force */
1653             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1654             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1655             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1656
1657             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1658             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1659             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1660
1661             /**************************
1662              * CALCULATE INTERACTIONS *
1663              **************************/
1664
1665             r11              = _mm_mul_pd(rsq11,rinv11);
1666
1667             /* EWALD ELECTROSTATICS */
1668
1669             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1670             ewrt             = _mm_mul_pd(r11,ewtabscale);
1671             ewitab           = _mm_cvttpd_epi32(ewrt);
1672 #ifdef __XOP__
1673             eweps            = _mm_frcz_pd(ewrt);
1674 #else
1675             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1676 #endif
1677             twoeweps         = _mm_add_pd(eweps,eweps);
1678             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1679                                          &ewtabF,&ewtabFn);
1680             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1681             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1682
1683             fscal            = felec;
1684
1685             /* Update vectorial force */
1686             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1687             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1688             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1689
1690             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1691             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1692             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1693
1694             /**************************
1695              * CALCULATE INTERACTIONS *
1696              **************************/
1697
1698             r12              = _mm_mul_pd(rsq12,rinv12);
1699
1700             /* EWALD ELECTROSTATICS */
1701
1702             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1703             ewrt             = _mm_mul_pd(r12,ewtabscale);
1704             ewitab           = _mm_cvttpd_epi32(ewrt);
1705 #ifdef __XOP__
1706             eweps            = _mm_frcz_pd(ewrt);
1707 #else
1708             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1709 #endif
1710             twoeweps         = _mm_add_pd(eweps,eweps);
1711             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1712                                          &ewtabF,&ewtabFn);
1713             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1714             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1715
1716             fscal            = felec;
1717
1718             /* Update vectorial force */
1719             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1720             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1721             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1722
1723             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1724             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1725             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1726
1727             /**************************
1728              * CALCULATE INTERACTIONS *
1729              **************************/
1730
1731             r13              = _mm_mul_pd(rsq13,rinv13);
1732
1733             /* EWALD ELECTROSTATICS */
1734
1735             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1736             ewrt             = _mm_mul_pd(r13,ewtabscale);
1737             ewitab           = _mm_cvttpd_epi32(ewrt);
1738 #ifdef __XOP__
1739             eweps            = _mm_frcz_pd(ewrt);
1740 #else
1741             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1742 #endif
1743             twoeweps         = _mm_add_pd(eweps,eweps);
1744             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1745                                          &ewtabF,&ewtabFn);
1746             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1747             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1748
1749             fscal            = felec;
1750
1751             /* Update vectorial force */
1752             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1753             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1754             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1755
1756             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1757             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1758             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1759
1760             /**************************
1761              * CALCULATE INTERACTIONS *
1762              **************************/
1763
1764             r21              = _mm_mul_pd(rsq21,rinv21);
1765
1766             /* EWALD ELECTROSTATICS */
1767
1768             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1769             ewrt             = _mm_mul_pd(r21,ewtabscale);
1770             ewitab           = _mm_cvttpd_epi32(ewrt);
1771 #ifdef __XOP__
1772             eweps            = _mm_frcz_pd(ewrt);
1773 #else
1774             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1775 #endif
1776             twoeweps         = _mm_add_pd(eweps,eweps);
1777             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1778                                          &ewtabF,&ewtabFn);
1779             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1780             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1781
1782             fscal            = felec;
1783
1784             /* Update vectorial force */
1785             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1786             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1787             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1788
1789             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1790             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1791             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1792
1793             /**************************
1794              * CALCULATE INTERACTIONS *
1795              **************************/
1796
1797             r22              = _mm_mul_pd(rsq22,rinv22);
1798
1799             /* EWALD ELECTROSTATICS */
1800
1801             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1802             ewrt             = _mm_mul_pd(r22,ewtabscale);
1803             ewitab           = _mm_cvttpd_epi32(ewrt);
1804 #ifdef __XOP__
1805             eweps            = _mm_frcz_pd(ewrt);
1806 #else
1807             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1808 #endif
1809             twoeweps         = _mm_add_pd(eweps,eweps);
1810             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1811                                          &ewtabF,&ewtabFn);
1812             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1813             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1814
1815             fscal            = felec;
1816
1817             /* Update vectorial force */
1818             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1819             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1820             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1821
1822             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1823             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1824             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1825
1826             /**************************
1827              * CALCULATE INTERACTIONS *
1828              **************************/
1829
1830             r23              = _mm_mul_pd(rsq23,rinv23);
1831
1832             /* EWALD ELECTROSTATICS */
1833
1834             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1835             ewrt             = _mm_mul_pd(r23,ewtabscale);
1836             ewitab           = _mm_cvttpd_epi32(ewrt);
1837 #ifdef __XOP__
1838             eweps            = _mm_frcz_pd(ewrt);
1839 #else
1840             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1841 #endif
1842             twoeweps         = _mm_add_pd(eweps,eweps);
1843             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1844                                          &ewtabF,&ewtabFn);
1845             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1846             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1847
1848             fscal            = felec;
1849
1850             /* Update vectorial force */
1851             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1852             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1853             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1854
1855             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1856             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1857             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1858
1859             /**************************
1860              * CALCULATE INTERACTIONS *
1861              **************************/
1862
1863             r31              = _mm_mul_pd(rsq31,rinv31);
1864
1865             /* EWALD ELECTROSTATICS */
1866
1867             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1868             ewrt             = _mm_mul_pd(r31,ewtabscale);
1869             ewitab           = _mm_cvttpd_epi32(ewrt);
1870 #ifdef __XOP__
1871             eweps            = _mm_frcz_pd(ewrt);
1872 #else
1873             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1874 #endif
1875             twoeweps         = _mm_add_pd(eweps,eweps);
1876             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1877                                          &ewtabF,&ewtabFn);
1878             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1879             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1880
1881             fscal            = felec;
1882
1883             /* Update vectorial force */
1884             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1885             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1886             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1887
1888             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1889             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1890             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1891
1892             /**************************
1893              * CALCULATE INTERACTIONS *
1894              **************************/
1895
1896             r32              = _mm_mul_pd(rsq32,rinv32);
1897
1898             /* EWALD ELECTROSTATICS */
1899
1900             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1901             ewrt             = _mm_mul_pd(r32,ewtabscale);
1902             ewitab           = _mm_cvttpd_epi32(ewrt);
1903 #ifdef __XOP__
1904             eweps            = _mm_frcz_pd(ewrt);
1905 #else
1906             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1907 #endif
1908             twoeweps         = _mm_add_pd(eweps,eweps);
1909             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1910                                          &ewtabF,&ewtabFn);
1911             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1912             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1913
1914             fscal            = felec;
1915
1916             /* Update vectorial force */
1917             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1918             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1919             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1920
1921             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1922             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1923             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1924
1925             /**************************
1926              * CALCULATE INTERACTIONS *
1927              **************************/
1928
1929             r33              = _mm_mul_pd(rsq33,rinv33);
1930
1931             /* EWALD ELECTROSTATICS */
1932
1933             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1934             ewrt             = _mm_mul_pd(r33,ewtabscale);
1935             ewitab           = _mm_cvttpd_epi32(ewrt);
1936 #ifdef __XOP__
1937             eweps            = _mm_frcz_pd(ewrt);
1938 #else
1939             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1940 #endif
1941             twoeweps         = _mm_add_pd(eweps,eweps);
1942             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1943                                          &ewtabF,&ewtabFn);
1944             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1945             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1946
1947             fscal            = felec;
1948
1949             /* Update vectorial force */
1950             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1951             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1952             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1953
1954             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1955             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1956             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1957
1958             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);
1959
1960             /* Inner loop uses 405 flops */
1961         }
1962
1963         if(jidx<j_index_end)
1964         {
1965
1966             jnrA             = jjnr[jidx];
1967             j_coord_offsetA  = DIM*jnrA;
1968
1969             /* load j atom coordinates */
1970             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1971                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1972                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1973
1974             /* Calculate displacement vector */
1975             dx00             = _mm_sub_pd(ix0,jx0);
1976             dy00             = _mm_sub_pd(iy0,jy0);
1977             dz00             = _mm_sub_pd(iz0,jz0);
1978             dx11             = _mm_sub_pd(ix1,jx1);
1979             dy11             = _mm_sub_pd(iy1,jy1);
1980             dz11             = _mm_sub_pd(iz1,jz1);
1981             dx12             = _mm_sub_pd(ix1,jx2);
1982             dy12             = _mm_sub_pd(iy1,jy2);
1983             dz12             = _mm_sub_pd(iz1,jz2);
1984             dx13             = _mm_sub_pd(ix1,jx3);
1985             dy13             = _mm_sub_pd(iy1,jy3);
1986             dz13             = _mm_sub_pd(iz1,jz3);
1987             dx21             = _mm_sub_pd(ix2,jx1);
1988             dy21             = _mm_sub_pd(iy2,jy1);
1989             dz21             = _mm_sub_pd(iz2,jz1);
1990             dx22             = _mm_sub_pd(ix2,jx2);
1991             dy22             = _mm_sub_pd(iy2,jy2);
1992             dz22             = _mm_sub_pd(iz2,jz2);
1993             dx23             = _mm_sub_pd(ix2,jx3);
1994             dy23             = _mm_sub_pd(iy2,jy3);
1995             dz23             = _mm_sub_pd(iz2,jz3);
1996             dx31             = _mm_sub_pd(ix3,jx1);
1997             dy31             = _mm_sub_pd(iy3,jy1);
1998             dz31             = _mm_sub_pd(iz3,jz1);
1999             dx32             = _mm_sub_pd(ix3,jx2);
2000             dy32             = _mm_sub_pd(iy3,jy2);
2001             dz32             = _mm_sub_pd(iz3,jz2);
2002             dx33             = _mm_sub_pd(ix3,jx3);
2003             dy33             = _mm_sub_pd(iy3,jy3);
2004             dz33             = _mm_sub_pd(iz3,jz3);
2005
2006             /* Calculate squared distance and things based on it */
2007             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2008             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2009             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2010             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
2011             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2012             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2013             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
2014             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
2015             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
2016             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
2017
2018             rinv00           = avx128fma_invsqrt_d(rsq00);
2019             rinv11           = avx128fma_invsqrt_d(rsq11);
2020             rinv12           = avx128fma_invsqrt_d(rsq12);
2021             rinv13           = avx128fma_invsqrt_d(rsq13);
2022             rinv21           = avx128fma_invsqrt_d(rsq21);
2023             rinv22           = avx128fma_invsqrt_d(rsq22);
2024             rinv23           = avx128fma_invsqrt_d(rsq23);
2025             rinv31           = avx128fma_invsqrt_d(rsq31);
2026             rinv32           = avx128fma_invsqrt_d(rsq32);
2027             rinv33           = avx128fma_invsqrt_d(rsq33);
2028
2029             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
2030             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
2031             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
2032             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
2033             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
2034             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
2035             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
2036             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
2037             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
2038
2039             fjx0             = _mm_setzero_pd();
2040             fjy0             = _mm_setzero_pd();
2041             fjz0             = _mm_setzero_pd();
2042             fjx1             = _mm_setzero_pd();
2043             fjy1             = _mm_setzero_pd();
2044             fjz1             = _mm_setzero_pd();
2045             fjx2             = _mm_setzero_pd();
2046             fjy2             = _mm_setzero_pd();
2047             fjz2             = _mm_setzero_pd();
2048             fjx3             = _mm_setzero_pd();
2049             fjy3             = _mm_setzero_pd();
2050             fjz3             = _mm_setzero_pd();
2051
2052             /**************************
2053              * CALCULATE INTERACTIONS *
2054              **************************/
2055
2056             r00              = _mm_mul_pd(rsq00,rinv00);
2057
2058             /* Calculate table index by multiplying r with table scale and truncate to integer */
2059             rt               = _mm_mul_pd(r00,vftabscale);
2060             vfitab           = _mm_cvttpd_epi32(rt);
2061 #ifdef __XOP__
2062             vfeps            = _mm_frcz_pd(rt);
2063 #else
2064             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2065 #endif
2066             twovfeps         = _mm_add_pd(vfeps,vfeps);
2067             vfitab           = _mm_slli_epi32(vfitab,3);
2068
2069             /* CUBIC SPLINE TABLE DISPERSION */
2070             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2071             F                = _mm_setzero_pd();
2072             GMX_MM_TRANSPOSE2_PD(Y,F);
2073             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2074             H                = _mm_setzero_pd();
2075             GMX_MM_TRANSPOSE2_PD(G,H);
2076             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
2077             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
2078             fvdw6            = _mm_mul_pd(c6_00,FF);
2079
2080             /* CUBIC SPLINE TABLE REPULSION */
2081             vfitab           = _mm_add_epi32(vfitab,ifour);
2082             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2083             F                = _mm_setzero_pd();
2084             GMX_MM_TRANSPOSE2_PD(Y,F);
2085             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2086             H                = _mm_setzero_pd();
2087             GMX_MM_TRANSPOSE2_PD(G,H);
2088             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
2089             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
2090             fvdw12           = _mm_mul_pd(c12_00,FF);
2091             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
2092
2093             fscal            = fvdw;
2094
2095             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2096
2097             /* Update vectorial force */
2098             fix0             = _mm_macc_pd(dx00,fscal,fix0);
2099             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
2100             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
2101
2102             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
2103             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
2104             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
2105
2106             /**************************
2107              * CALCULATE INTERACTIONS *
2108              **************************/
2109
2110             r11              = _mm_mul_pd(rsq11,rinv11);
2111
2112             /* EWALD ELECTROSTATICS */
2113
2114             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2115             ewrt             = _mm_mul_pd(r11,ewtabscale);
2116             ewitab           = _mm_cvttpd_epi32(ewrt);
2117 #ifdef __XOP__
2118             eweps            = _mm_frcz_pd(ewrt);
2119 #else
2120             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2121 #endif
2122             twoeweps         = _mm_add_pd(eweps,eweps);
2123             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2124             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2125             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2126
2127             fscal            = felec;
2128
2129             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2130
2131             /* Update vectorial force */
2132             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2133             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2134             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2135
2136             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2137             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2138             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2139
2140             /**************************
2141              * CALCULATE INTERACTIONS *
2142              **************************/
2143
2144             r12              = _mm_mul_pd(rsq12,rinv12);
2145
2146             /* EWALD ELECTROSTATICS */
2147
2148             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2149             ewrt             = _mm_mul_pd(r12,ewtabscale);
2150             ewitab           = _mm_cvttpd_epi32(ewrt);
2151 #ifdef __XOP__
2152             eweps            = _mm_frcz_pd(ewrt);
2153 #else
2154             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2155 #endif
2156             twoeweps         = _mm_add_pd(eweps,eweps);
2157             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2158             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2159             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2160
2161             fscal            = felec;
2162
2163             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2164
2165             /* Update vectorial force */
2166             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2167             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2168             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2169
2170             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2171             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2172             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2173
2174             /**************************
2175              * CALCULATE INTERACTIONS *
2176              **************************/
2177
2178             r13              = _mm_mul_pd(rsq13,rinv13);
2179
2180             /* EWALD ELECTROSTATICS */
2181
2182             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2183             ewrt             = _mm_mul_pd(r13,ewtabscale);
2184             ewitab           = _mm_cvttpd_epi32(ewrt);
2185 #ifdef __XOP__
2186             eweps            = _mm_frcz_pd(ewrt);
2187 #else
2188             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2189 #endif
2190             twoeweps         = _mm_add_pd(eweps,eweps);
2191             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2192             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2193             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
2194
2195             fscal            = felec;
2196
2197             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2198
2199             /* Update vectorial force */
2200             fix1             = _mm_macc_pd(dx13,fscal,fix1);
2201             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
2202             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
2203
2204             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
2205             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
2206             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
2207
2208             /**************************
2209              * CALCULATE INTERACTIONS *
2210              **************************/
2211
2212             r21              = _mm_mul_pd(rsq21,rinv21);
2213
2214             /* EWALD ELECTROSTATICS */
2215
2216             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2217             ewrt             = _mm_mul_pd(r21,ewtabscale);
2218             ewitab           = _mm_cvttpd_epi32(ewrt);
2219 #ifdef __XOP__
2220             eweps            = _mm_frcz_pd(ewrt);
2221 #else
2222             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2223 #endif
2224             twoeweps         = _mm_add_pd(eweps,eweps);
2225             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2226             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2227             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2228
2229             fscal            = felec;
2230
2231             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2232
2233             /* Update vectorial force */
2234             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2235             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2236             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2237
2238             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2239             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2240             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2241
2242             /**************************
2243              * CALCULATE INTERACTIONS *
2244              **************************/
2245
2246             r22              = _mm_mul_pd(rsq22,rinv22);
2247
2248             /* EWALD ELECTROSTATICS */
2249
2250             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2251             ewrt             = _mm_mul_pd(r22,ewtabscale);
2252             ewitab           = _mm_cvttpd_epi32(ewrt);
2253 #ifdef __XOP__
2254             eweps            = _mm_frcz_pd(ewrt);
2255 #else
2256             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2257 #endif
2258             twoeweps         = _mm_add_pd(eweps,eweps);
2259             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2260             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2261             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2262
2263             fscal            = felec;
2264
2265             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2266
2267             /* Update vectorial force */
2268             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2269             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2270             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2271
2272             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2273             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2274             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2275
2276             /**************************
2277              * CALCULATE INTERACTIONS *
2278              **************************/
2279
2280             r23              = _mm_mul_pd(rsq23,rinv23);
2281
2282             /* EWALD ELECTROSTATICS */
2283
2284             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2285             ewrt             = _mm_mul_pd(r23,ewtabscale);
2286             ewitab           = _mm_cvttpd_epi32(ewrt);
2287 #ifdef __XOP__
2288             eweps            = _mm_frcz_pd(ewrt);
2289 #else
2290             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2291 #endif
2292             twoeweps         = _mm_add_pd(eweps,eweps);
2293             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2294             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2295             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
2296
2297             fscal            = felec;
2298
2299             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2300
2301             /* Update vectorial force */
2302             fix2             = _mm_macc_pd(dx23,fscal,fix2);
2303             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
2304             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
2305
2306             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
2307             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
2308             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
2309
2310             /**************************
2311              * CALCULATE INTERACTIONS *
2312              **************************/
2313
2314             r31              = _mm_mul_pd(rsq31,rinv31);
2315
2316             /* EWALD ELECTROSTATICS */
2317
2318             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2319             ewrt             = _mm_mul_pd(r31,ewtabscale);
2320             ewitab           = _mm_cvttpd_epi32(ewrt);
2321 #ifdef __XOP__
2322             eweps            = _mm_frcz_pd(ewrt);
2323 #else
2324             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2325 #endif
2326             twoeweps         = _mm_add_pd(eweps,eweps);
2327             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2328             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2329             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2330
2331             fscal            = felec;
2332
2333             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2334
2335             /* Update vectorial force */
2336             fix3             = _mm_macc_pd(dx31,fscal,fix3);
2337             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
2338             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
2339
2340             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
2341             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
2342             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
2343
2344             /**************************
2345              * CALCULATE INTERACTIONS *
2346              **************************/
2347
2348             r32              = _mm_mul_pd(rsq32,rinv32);
2349
2350             /* EWALD ELECTROSTATICS */
2351
2352             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2353             ewrt             = _mm_mul_pd(r32,ewtabscale);
2354             ewitab           = _mm_cvttpd_epi32(ewrt);
2355 #ifdef __XOP__
2356             eweps            = _mm_frcz_pd(ewrt);
2357 #else
2358             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2359 #endif
2360             twoeweps         = _mm_add_pd(eweps,eweps);
2361             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2362             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2363             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2364
2365             fscal            = felec;
2366
2367             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2368
2369             /* Update vectorial force */
2370             fix3             = _mm_macc_pd(dx32,fscal,fix3);
2371             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
2372             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
2373
2374             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
2375             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
2376             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
2377
2378             /**************************
2379              * CALCULATE INTERACTIONS *
2380              **************************/
2381
2382             r33              = _mm_mul_pd(rsq33,rinv33);
2383
2384             /* EWALD ELECTROSTATICS */
2385
2386             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2387             ewrt             = _mm_mul_pd(r33,ewtabscale);
2388             ewitab           = _mm_cvttpd_epi32(ewrt);
2389 #ifdef __XOP__
2390             eweps            = _mm_frcz_pd(ewrt);
2391 #else
2392             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2393 #endif
2394             twoeweps         = _mm_add_pd(eweps,eweps);
2395             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2396             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2397             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2398
2399             fscal            = felec;
2400
2401             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2402
2403             /* Update vectorial force */
2404             fix3             = _mm_macc_pd(dx33,fscal,fix3);
2405             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
2406             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
2407
2408             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
2409             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
2410             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
2411
2412             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2413
2414             /* Inner loop uses 405 flops */
2415         }
2416
2417         /* End of innermost loop */
2418
2419         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2420                                               f+i_coord_offset,fshift+i_shift_offset);
2421
2422         /* Increment number of inner iterations */
2423         inneriter                  += j_index_end - j_index_start;
2424
2425         /* Outer loop uses 24 flops */
2426     }
2427
2428     /* Increment number of outer iterations */
2429     outeriter        += nri;
2430
2431     /* Update outer/inner flops */
2432
2433     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*405);
2434 }