src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJ_GeomW3W3_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3W3_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW3W3_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     int              vdwjidx1A,vdwjidx1B;
  91     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  92     int              vdwjidx2A,vdwjidx2B;
  93     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  94     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  96     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  97     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  99     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 100     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 102     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 103     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 104     real             *charge;
 105     int              nvdwtype;
 106     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 107     int              *vdwtype;
 108     real             *vdwparam;
 109     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 110     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 111     __m128i          ewitab;
 112     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     real             *ewtab;
 114     __m128d          dummy_mask,cutoff_mask;
 115     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 116     __m128d          one     = _mm_set1_pd(1.0);
 117     __m128d          two     = _mm_set1_pd(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm_set1_pd(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134
 135     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 136     ewtab            = fr->ic->tabq_coul_FDV0;
 137     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 138     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 143     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 144     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 145     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 146
 147     jq0              = _mm_set1_pd(charge[inr+0]);
 148     jq1              = _mm_set1_pd(charge[inr+1]);
 149     jq2              = _mm_set1_pd(charge[inr+2]);
 150     vdwjidx0A        = 2*vdwtype[inr+0];
 151     qq00             = _mm_mul_pd(iq0,jq0);
 152     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 153     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 154     qq01             = _mm_mul_pd(iq0,jq1);
 155     qq02             = _mm_mul_pd(iq0,jq2);
 156     qq10             = _mm_mul_pd(iq1,jq0);
 157     qq11             = _mm_mul_pd(iq1,jq1);
 158     qq12             = _mm_mul_pd(iq1,jq2);
 159     qq20             = _mm_mul_pd(iq2,jq0);
 160     qq21             = _mm_mul_pd(iq2,jq1);
 161     qq22             = _mm_mul_pd(iq2,jq2);
 162
 163     /* Avoid stupid compiler warnings */
 164     jnrA = jnrB = 0;
 165     j_coord_offsetA = 0;
 166     j_coord_offsetB = 0;
 167
 168     outeriter        = 0;
 169     inneriter        = 0;
 170
 171     /* Start outer loop over neighborlists */
 172     for(iidx=0; iidx<nri; iidx++)
 173     {
 174         /* Load shift vector for this list */
 175         i_shift_offset   = DIM*shiftidx[iidx];
 176
 177         /* Load limits for loop over neighbors */
 178         j_index_start    = jindex[iidx];
 179         j_index_end      = jindex[iidx+1];
 180
 181         /* Get outer coordinate index */
 182         inr              = iinr[iidx];
 183         i_coord_offset   = DIM*inr;
 184
 185         /* Load i particle coords and add shift vector */
 186         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 187                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 188
 189         fix0             = _mm_setzero_pd();
 190         fiy0             = _mm_setzero_pd();
 191         fiz0             = _mm_setzero_pd();
 192         fix1             = _mm_setzero_pd();
 193         fiy1             = _mm_setzero_pd();
 194         fiz1             = _mm_setzero_pd();
 195         fix2             = _mm_setzero_pd();
 196         fiy2             = _mm_setzero_pd();
 197         fiz2             = _mm_setzero_pd();
 198
 199         /* Reset potential sums */
 200         velecsum         = _mm_setzero_pd();
 201         vvdwsum          = _mm_setzero_pd();
 202
 203         /* Start inner kernel loop */
 204         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 205         {
 206
 207             /* Get j neighbor index, and coordinate index */
 208             jnrA             = jjnr[jidx];
 209             jnrB             = jjnr[jidx+1];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212
 213             /* load j atom coordinates */
 214             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 215                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm_sub_pd(ix0,jx0);
 219             dy00             = _mm_sub_pd(iy0,jy0);
 220             dz00             = _mm_sub_pd(iz0,jz0);
 221             dx01             = _mm_sub_pd(ix0,jx1);
 222             dy01             = _mm_sub_pd(iy0,jy1);
 223             dz01             = _mm_sub_pd(iz0,jz1);
 224             dx02             = _mm_sub_pd(ix0,jx2);
 225             dy02             = _mm_sub_pd(iy0,jy2);
 226             dz02             = _mm_sub_pd(iz0,jz2);
 227             dx10             = _mm_sub_pd(ix1,jx0);
 228             dy10             = _mm_sub_pd(iy1,jy0);
 229             dz10             = _mm_sub_pd(iz1,jz0);
 230             dx11             = _mm_sub_pd(ix1,jx1);
 231             dy11             = _mm_sub_pd(iy1,jy1);
 232             dz11             = _mm_sub_pd(iz1,jz1);
 233             dx12             = _mm_sub_pd(ix1,jx2);
 234             dy12             = _mm_sub_pd(iy1,jy2);
 235             dz12             = _mm_sub_pd(iz1,jz2);
 236             dx20             = _mm_sub_pd(ix2,jx0);
 237             dy20             = _mm_sub_pd(iy2,jy0);
 238             dz20             = _mm_sub_pd(iz2,jz0);
 239             dx21             = _mm_sub_pd(ix2,jx1);
 240             dy21             = _mm_sub_pd(iy2,jy1);
 241             dz21             = _mm_sub_pd(iz2,jz1);
 242             dx22             = _mm_sub_pd(ix2,jx2);
 243             dy22             = _mm_sub_pd(iy2,jy2);
 244             dz22             = _mm_sub_pd(iz2,jz2);
 245
 246             /* Calculate squared distance and things based on it */
 247             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 248             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 249             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 250             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 251             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 252             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 253             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 254             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 255             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 256
 257             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 258             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 259             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 260             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 261             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 262             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 263             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 264             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 265             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 266
 267             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 268             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 269             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 270             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 271             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 272             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 273             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 274             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 275             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 276
 277             fjx0             = _mm_setzero_pd();
 278             fjy0             = _mm_setzero_pd();
 279             fjz0             = _mm_setzero_pd();
 280             fjx1             = _mm_setzero_pd();
 281             fjy1             = _mm_setzero_pd();
 282             fjz1             = _mm_setzero_pd();
 283             fjx2             = _mm_setzero_pd();
 284             fjy2             = _mm_setzero_pd();
 285             fjz2             = _mm_setzero_pd();
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             r00              = _mm_mul_pd(rsq00,rinv00);
 292
 293             /* EWALD ELECTROSTATICS */
 294
 295             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 296             ewrt             = _mm_mul_pd(r00,ewtabscale);
 297             ewitab           = _mm_cvttpd_epi32(ewrt);
 298 #ifdef __XOP__
 299             eweps            = _mm_frcz_pd(ewrt);
 300 #else
 301             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 302 #endif
 303             twoeweps         = _mm_add_pd(eweps,eweps);
 304             ewitab           = _mm_slli_epi32(ewitab,2);
 305             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 306             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 307             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 308             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 309             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 310             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 311             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 312             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 313             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 314             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 315
 316             /* LENNARD-JONES DISPERSION/REPULSION */
 317
 318             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 319             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 320             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 321             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 322             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velecsum         = _mm_add_pd(velecsum,velec);
 326             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 327
 328             fscal            = _mm_add_pd(felec,fvdw);
 329
 330             /* Update vectorial force */
 331             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 332             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 333             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 334
 335             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 336             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 337             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             r01              = _mm_mul_pd(rsq01,rinv01);
 344
 345             /* EWALD ELECTROSTATICS */
 346
 347             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 348             ewrt             = _mm_mul_pd(r01,ewtabscale);
 349             ewitab           = _mm_cvttpd_epi32(ewrt);
 350 #ifdef __XOP__
 351             eweps            = _mm_frcz_pd(ewrt);
 352 #else
 353             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 354 #endif
 355             twoeweps         = _mm_add_pd(eweps,eweps);
 356             ewitab           = _mm_slli_epi32(ewitab,2);
 357             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 358             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 359             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 360             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 361             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 362             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 363             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 364             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 365             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 366             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velecsum         = _mm_add_pd(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             /* Update vectorial force */
 374             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 375             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 376             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 377
 378             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 379             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 380             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 381
 382             /**************************
 383              * CALCULATE INTERACTIONS *
 384              **************************/
 385
 386             r02              = _mm_mul_pd(rsq02,rinv02);
 387
 388             /* EWALD ELECTROSTATICS */
 389
 390             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 391             ewrt             = _mm_mul_pd(r02,ewtabscale);
 392             ewitab           = _mm_cvttpd_epi32(ewrt);
 393 #ifdef __XOP__
 394             eweps            = _mm_frcz_pd(ewrt);
 395 #else
 396             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 397 #endif
 398             twoeweps         = _mm_add_pd(eweps,eweps);
 399             ewitab           = _mm_slli_epi32(ewitab,2);
 400             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 401             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 402             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 403             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 404             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 405             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 406             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 407             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 408             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 409             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velecsum         = _mm_add_pd(velecsum,velec);
 413
 414             fscal            = felec;
 415
 416             /* Update vectorial force */
 417             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 418             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 419             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 420
 421             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 422             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 423             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             r10              = _mm_mul_pd(rsq10,rinv10);
 430
 431             /* EWALD ELECTROSTATICS */
 432
 433             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 434             ewrt             = _mm_mul_pd(r10,ewtabscale);
 435             ewitab           = _mm_cvttpd_epi32(ewrt);
 436 #ifdef __XOP__
 437             eweps            = _mm_frcz_pd(ewrt);
 438 #else
 439             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 440 #endif
 441             twoeweps         = _mm_add_pd(eweps,eweps);
 442             ewitab           = _mm_slli_epi32(ewitab,2);
 443             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 444             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 445             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 446             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 447             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 448             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 449             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 450             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 451             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 452             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             velecsum         = _mm_add_pd(velecsum,velec);
 456
 457             fscal            = felec;
 458
 459             /* Update vectorial force */
 460             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 461             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 462             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 463
 464             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 465             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 466             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 467
 468             /**************************
 469              * CALCULATE INTERACTIONS *
 470              **************************/
 471
 472             r11              = _mm_mul_pd(rsq11,rinv11);
 473
 474             /* EWALD ELECTROSTATICS */
 475
 476             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 477             ewrt             = _mm_mul_pd(r11,ewtabscale);
 478             ewitab           = _mm_cvttpd_epi32(ewrt);
 479 #ifdef __XOP__
 480             eweps            = _mm_frcz_pd(ewrt);
 481 #else
 482             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 483 #endif
 484             twoeweps         = _mm_add_pd(eweps,eweps);
 485             ewitab           = _mm_slli_epi32(ewitab,2);
 486             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 487             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 488             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 489             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 490             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 491             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 492             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 493             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 494             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 495             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 496
 497             /* Update potential sum for this i atom from the interaction with this j atom. */
 498             velecsum         = _mm_add_pd(velecsum,velec);
 499
 500             fscal            = felec;
 501
 502             /* Update vectorial force */
 503             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 504             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 505             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 506
 507             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 508             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 509             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             r12              = _mm_mul_pd(rsq12,rinv12);
 516
 517             /* EWALD ELECTROSTATICS */
 518
 519             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 520             ewrt             = _mm_mul_pd(r12,ewtabscale);
 521             ewitab           = _mm_cvttpd_epi32(ewrt);
 522 #ifdef __XOP__
 523             eweps            = _mm_frcz_pd(ewrt);
 524 #else
 525             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 526 #endif
 527             twoeweps         = _mm_add_pd(eweps,eweps);
 528             ewitab           = _mm_slli_epi32(ewitab,2);
 529             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 530             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 531             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 532             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 533             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 534             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 535             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 536             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 537             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 538             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             velecsum         = _mm_add_pd(velecsum,velec);
 542
 543             fscal            = felec;
 544
 545             /* Update vectorial force */
 546             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 547             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 548             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 549
 550             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 551             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 552             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             r20              = _mm_mul_pd(rsq20,rinv20);
 559
 560             /* EWALD ELECTROSTATICS */
 561
 562             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 563             ewrt             = _mm_mul_pd(r20,ewtabscale);
 564             ewitab           = _mm_cvttpd_epi32(ewrt);
 565 #ifdef __XOP__
 566             eweps            = _mm_frcz_pd(ewrt);
 567 #else
 568             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 569 #endif
 570             twoeweps         = _mm_add_pd(eweps,eweps);
 571             ewitab           = _mm_slli_epi32(ewitab,2);
 572             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 573             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 574             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 575             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 576             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 577             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 578             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 579             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 580             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 581             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 582
 583             /* Update potential sum for this i atom from the interaction with this j atom. */
 584             velecsum         = _mm_add_pd(velecsum,velec);
 585
 586             fscal            = felec;
 587
 588             /* Update vectorial force */
 589             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 590             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 591             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 592
 593             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 594             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 595             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 596
 597             /**************************
 598              * CALCULATE INTERACTIONS *
 599              **************************/
 600
 601             r21              = _mm_mul_pd(rsq21,rinv21);
 602
 603             /* EWALD ELECTROSTATICS */
 604
 605             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 606             ewrt             = _mm_mul_pd(r21,ewtabscale);
 607             ewitab           = _mm_cvttpd_epi32(ewrt);
 608 #ifdef __XOP__
 609             eweps            = _mm_frcz_pd(ewrt);
 610 #else
 611             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 612 #endif
 613             twoeweps         = _mm_add_pd(eweps,eweps);
 614             ewitab           = _mm_slli_epi32(ewitab,2);
 615             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 616             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 617             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 618             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 619             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 620             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 621             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 622             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 623             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 624             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velecsum         = _mm_add_pd(velecsum,velec);
 628
 629             fscal            = felec;
 630
 631             /* Update vectorial force */
 632             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 633             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 634             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 635
 636             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 637             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 638             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 639
 640             /**************************
 641              * CALCULATE INTERACTIONS *
 642              **************************/
 643
 644             r22              = _mm_mul_pd(rsq22,rinv22);
 645
 646             /* EWALD ELECTROSTATICS */
 647
 648             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 649             ewrt             = _mm_mul_pd(r22,ewtabscale);
 650             ewitab           = _mm_cvttpd_epi32(ewrt);
 651 #ifdef __XOP__
 652             eweps            = _mm_frcz_pd(ewrt);
 653 #else
 654             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 655 #endif
 656             twoeweps         = _mm_add_pd(eweps,eweps);
 657             ewitab           = _mm_slli_epi32(ewitab,2);
 658             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 659             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 660             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 661             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 662             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 663             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 664             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 665             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 666             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 667             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 668
 669             /* Update potential sum for this i atom from the interaction with this j atom. */
 670             velecsum         = _mm_add_pd(velecsum,velec);
 671
 672             fscal            = felec;
 673
 674             /* Update vectorial force */
 675             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 676             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 677             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 678
 679             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 680             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 681             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 682
 683             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 684
 685             /* Inner loop uses 408 flops */
 686         }
 687
 688         if(jidx<j_index_end)
 689         {
 690
 691             jnrA             = jjnr[jidx];
 692             j_coord_offsetA  = DIM*jnrA;
 693
 694             /* load j atom coordinates */
 695             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 696                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 697
 698             /* Calculate displacement vector */
 699             dx00             = _mm_sub_pd(ix0,jx0);
 700             dy00             = _mm_sub_pd(iy0,jy0);
 701             dz00             = _mm_sub_pd(iz0,jz0);
 702             dx01             = _mm_sub_pd(ix0,jx1);
 703             dy01             = _mm_sub_pd(iy0,jy1);
 704             dz01             = _mm_sub_pd(iz0,jz1);
 705             dx02             = _mm_sub_pd(ix0,jx2);
 706             dy02             = _mm_sub_pd(iy0,jy2);
 707             dz02             = _mm_sub_pd(iz0,jz2);
 708             dx10             = _mm_sub_pd(ix1,jx0);
 709             dy10             = _mm_sub_pd(iy1,jy0);
 710             dz10             = _mm_sub_pd(iz1,jz0);
 711             dx11             = _mm_sub_pd(ix1,jx1);
 712             dy11             = _mm_sub_pd(iy1,jy1);
 713             dz11             = _mm_sub_pd(iz1,jz1);
 714             dx12             = _mm_sub_pd(ix1,jx2);
 715             dy12             = _mm_sub_pd(iy1,jy2);
 716             dz12             = _mm_sub_pd(iz1,jz2);
 717             dx20             = _mm_sub_pd(ix2,jx0);
 718             dy20             = _mm_sub_pd(iy2,jy0);
 719             dz20             = _mm_sub_pd(iz2,jz0);
 720             dx21             = _mm_sub_pd(ix2,jx1);
 721             dy21             = _mm_sub_pd(iy2,jy1);
 722             dz21             = _mm_sub_pd(iz2,jz1);
 723             dx22             = _mm_sub_pd(ix2,jx2);
 724             dy22             = _mm_sub_pd(iy2,jy2);
 725             dz22             = _mm_sub_pd(iz2,jz2);
 726
 727             /* Calculate squared distance and things based on it */
 728             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 729             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 730             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 731             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 732             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 733             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 734             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 735             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 736             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 737
 738             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 739             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 740             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 741             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 742             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 743             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 744             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 745             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 746             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 747
 748             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 749             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 750             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 751             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 752             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 753             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 754             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 755             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 756             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 757
 758             fjx0             = _mm_setzero_pd();
 759             fjy0             = _mm_setzero_pd();
 760             fjz0             = _mm_setzero_pd();
 761             fjx1             = _mm_setzero_pd();
 762             fjy1             = _mm_setzero_pd();
 763             fjz1             = _mm_setzero_pd();
 764             fjx2             = _mm_setzero_pd();
 765             fjy2             = _mm_setzero_pd();
 766             fjz2             = _mm_setzero_pd();
 767
 768             /**************************
 769              * CALCULATE INTERACTIONS *
 770              **************************/
 771
 772             r00              = _mm_mul_pd(rsq00,rinv00);
 773
 774             /* EWALD ELECTROSTATICS */
 775
 776             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 777             ewrt             = _mm_mul_pd(r00,ewtabscale);
 778             ewitab           = _mm_cvttpd_epi32(ewrt);
 779 #ifdef __XOP__
 780             eweps            = _mm_frcz_pd(ewrt);
 781 #else
 782             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 783 #endif
 784             twoeweps         = _mm_add_pd(eweps,eweps);
 785             ewitab           = _mm_slli_epi32(ewitab,2);
 786             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 787             ewtabD           = _mm_setzero_pd();
 788             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 789             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 790             ewtabFn          = _mm_setzero_pd();
 791             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 792             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 793             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 794             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 795             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 796
 797             /* LENNARD-JONES DISPERSION/REPULSION */
 798
 799             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 800             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 801             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 802             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 803             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 804
 805             /* Update potential sum for this i atom from the interaction with this j atom. */
 806             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 807             velecsum         = _mm_add_pd(velecsum,velec);
 808             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 809             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 810
 811             fscal            = _mm_add_pd(felec,fvdw);
 812
 813             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 814
 815             /* Update vectorial force */
 816             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 817             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 818             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 819
 820             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 821             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 822             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 823
 824             /**************************
 825              * CALCULATE INTERACTIONS *
 826              **************************/
 827
 828             r01              = _mm_mul_pd(rsq01,rinv01);
 829
 830             /* EWALD ELECTROSTATICS */
 831
 832             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 833             ewrt             = _mm_mul_pd(r01,ewtabscale);
 834             ewitab           = _mm_cvttpd_epi32(ewrt);
 835 #ifdef __XOP__
 836             eweps            = _mm_frcz_pd(ewrt);
 837 #else
 838             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 839 #endif
 840             twoeweps         = _mm_add_pd(eweps,eweps);
 841             ewitab           = _mm_slli_epi32(ewitab,2);
 842             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 843             ewtabD           = _mm_setzero_pd();
 844             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 845             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 846             ewtabFn          = _mm_setzero_pd();
 847             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 848             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 849             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 850             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 851             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 852
 853             /* Update potential sum for this i atom from the interaction with this j atom. */
 854             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 855             velecsum         = _mm_add_pd(velecsum,velec);
 856
 857             fscal            = felec;
 858
 859             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 860
 861             /* Update vectorial force */
 862             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 863             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 864             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 865
 866             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 867             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 868             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 869
 870             /**************************
 871              * CALCULATE INTERACTIONS *
 872              **************************/
 873
 874             r02              = _mm_mul_pd(rsq02,rinv02);
 875
 876             /* EWALD ELECTROSTATICS */
 877
 878             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 879             ewrt             = _mm_mul_pd(r02,ewtabscale);
 880             ewitab           = _mm_cvttpd_epi32(ewrt);
 881 #ifdef __XOP__
 882             eweps            = _mm_frcz_pd(ewrt);
 883 #else
 884             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 885 #endif
 886             twoeweps         = _mm_add_pd(eweps,eweps);
 887             ewitab           = _mm_slli_epi32(ewitab,2);
 888             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 889             ewtabD           = _mm_setzero_pd();
 890             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 891             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 892             ewtabFn          = _mm_setzero_pd();
 893             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 894             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 895             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 896             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 897             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 898
 899             /* Update potential sum for this i atom from the interaction with this j atom. */
 900             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 901             velecsum         = _mm_add_pd(velecsum,velec);
 902
 903             fscal            = felec;
 904
 905             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 906
 907             /* Update vectorial force */
 908             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 909             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 910             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 911
 912             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 913             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 914             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 915
 916             /**************************
 917              * CALCULATE INTERACTIONS *
 918              **************************/
 919
 920             r10              = _mm_mul_pd(rsq10,rinv10);
 921
 922             /* EWALD ELECTROSTATICS */
 923
 924             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 925             ewrt             = _mm_mul_pd(r10,ewtabscale);
 926             ewitab           = _mm_cvttpd_epi32(ewrt);
 927 #ifdef __XOP__
 928             eweps            = _mm_frcz_pd(ewrt);
 929 #else
 930             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 931 #endif
 932             twoeweps         = _mm_add_pd(eweps,eweps);
 933             ewitab           = _mm_slli_epi32(ewitab,2);
 934             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 935             ewtabD           = _mm_setzero_pd();
 936             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 937             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 938             ewtabFn          = _mm_setzero_pd();
 939             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 940             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 941             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 942             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 943             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 944
 945             /* Update potential sum for this i atom from the interaction with this j atom. */
 946             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 947             velecsum         = _mm_add_pd(velecsum,velec);
 948
 949             fscal            = felec;
 950
 951             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 952
 953             /* Update vectorial force */
 954             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 955             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 956             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 957
 958             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 959             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 960             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 961
 962             /**************************
 963              * CALCULATE INTERACTIONS *
 964              **************************/
 965
 966             r11              = _mm_mul_pd(rsq11,rinv11);
 967
 968             /* EWALD ELECTROSTATICS */
 969
 970             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 971             ewrt             = _mm_mul_pd(r11,ewtabscale);
 972             ewitab           = _mm_cvttpd_epi32(ewrt);
 973 #ifdef __XOP__
 974             eweps            = _mm_frcz_pd(ewrt);
 975 #else
 976             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 977 #endif
 978             twoeweps         = _mm_add_pd(eweps,eweps);
 979             ewitab           = _mm_slli_epi32(ewitab,2);
 980             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 981             ewtabD           = _mm_setzero_pd();
 982             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 983             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 984             ewtabFn          = _mm_setzero_pd();
 985             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 986             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 987             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 988             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 989             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 990
 991             /* Update potential sum for this i atom from the interaction with this j atom. */
 992             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 993             velecsum         = _mm_add_pd(velecsum,velec);
 994
 995             fscal            = felec;
 996
 997             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 998
 999             /* Update vectorial force */
1000             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1001             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1002             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1003
1004             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1005             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1006             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             r12              = _mm_mul_pd(rsq12,rinv12);
1013
1014             /* EWALD ELECTROSTATICS */
1015
1016             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1017             ewrt             = _mm_mul_pd(r12,ewtabscale);
1018             ewitab           = _mm_cvttpd_epi32(ewrt);
1019 #ifdef __XOP__
1020             eweps            = _mm_frcz_pd(ewrt);
1021 #else
1022             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1023 #endif
1024             twoeweps         = _mm_add_pd(eweps,eweps);
1025             ewitab           = _mm_slli_epi32(ewitab,2);
1026             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1027             ewtabD           = _mm_setzero_pd();
1028             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1029             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1030             ewtabFn          = _mm_setzero_pd();
1031             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1032             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1033             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1034             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
1035             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1036
1037             /* Update potential sum for this i atom from the interaction with this j atom. */
1038             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1039             velecsum         = _mm_add_pd(velecsum,velec);
1040
1041             fscal            = felec;
1042
1043             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1044
1045             /* Update vectorial force */
1046             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1047             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1048             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1049
1050             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1051             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1052             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             r20              = _mm_mul_pd(rsq20,rinv20);
1059
1060             /* EWALD ELECTROSTATICS */
1061
1062             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1063             ewrt             = _mm_mul_pd(r20,ewtabscale);
1064             ewitab           = _mm_cvttpd_epi32(ewrt);
1065 #ifdef __XOP__
1066             eweps            = _mm_frcz_pd(ewrt);
1067 #else
1068             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1069 #endif
1070             twoeweps         = _mm_add_pd(eweps,eweps);
1071             ewitab           = _mm_slli_epi32(ewitab,2);
1072             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1073             ewtabD           = _mm_setzero_pd();
1074             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1075             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1076             ewtabFn          = _mm_setzero_pd();
1077             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1078             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1079             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1080             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1081             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1082
1083             /* Update potential sum for this i atom from the interaction with this j atom. */
1084             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1085             velecsum         = _mm_add_pd(velecsum,velec);
1086
1087             fscal            = felec;
1088
1089             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1090
1091             /* Update vectorial force */
1092             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1093             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1094             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1095
1096             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1097             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1098             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1099
1100             /**************************
1101              * CALCULATE INTERACTIONS *
1102              **************************/
1103
1104             r21              = _mm_mul_pd(rsq21,rinv21);
1105
1106             /* EWALD ELECTROSTATICS */
1107
1108             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1109             ewrt             = _mm_mul_pd(r21,ewtabscale);
1110             ewitab           = _mm_cvttpd_epi32(ewrt);
1111 #ifdef __XOP__
1112             eweps            = _mm_frcz_pd(ewrt);
1113 #else
1114             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1115 #endif
1116             twoeweps         = _mm_add_pd(eweps,eweps);
1117             ewitab           = _mm_slli_epi32(ewitab,2);
1118             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1119             ewtabD           = _mm_setzero_pd();
1120             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1121             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1122             ewtabFn          = _mm_setzero_pd();
1123             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1124             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1125             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1126             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1127             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1128
1129             /* Update potential sum for this i atom from the interaction with this j atom. */
1130             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1131             velecsum         = _mm_add_pd(velecsum,velec);
1132
1133             fscal            = felec;
1134
1135             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1136
1137             /* Update vectorial force */
1138             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1139             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1140             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1141
1142             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1143             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1144             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1145
1146             /**************************
1147              * CALCULATE INTERACTIONS *
1148              **************************/
1149
1150             r22              = _mm_mul_pd(rsq22,rinv22);
1151
1152             /* EWALD ELECTROSTATICS */
1153
1154             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1155             ewrt             = _mm_mul_pd(r22,ewtabscale);
1156             ewitab           = _mm_cvttpd_epi32(ewrt);
1157 #ifdef __XOP__
1158             eweps            = _mm_frcz_pd(ewrt);
1159 #else
1160             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1161 #endif
1162             twoeweps         = _mm_add_pd(eweps,eweps);
1163             ewitab           = _mm_slli_epi32(ewitab,2);
1164             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1165             ewtabD           = _mm_setzero_pd();
1166             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1167             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1168             ewtabFn          = _mm_setzero_pd();
1169             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1170             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1171             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1172             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1173             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1174
1175             /* Update potential sum for this i atom from the interaction with this j atom. */
1176             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1177             velecsum         = _mm_add_pd(velecsum,velec);
1178
1179             fscal            = felec;
1180
1181             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1182
1183             /* Update vectorial force */
1184             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1185             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1186             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1187
1188             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1189             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1190             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1191
1192             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1193
1194             /* Inner loop uses 408 flops */
1195         }
1196
1197         /* End of innermost loop */
1198
1199         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1200                                               f+i_coord_offset,fshift+i_shift_offset);
1201
1202         ggid                        = gid[iidx];
1203         /* Update potential energies */
1204         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1205         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1206
1207         /* Increment number of inner iterations */
1208         inneriter                  += j_index_end - j_index_start;
1209
1210         /* Outer loop uses 20 flops */
1211     }
1212
1213     /* Increment number of outer iterations */
1214     outeriter        += nri;
1215
1216     /* Update outer/inner flops */
1217
1218     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*408);
1219 }
1220 /*
1221  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3W3_F_avx_128_fma_double
1222  * Electrostatics interaction: Ewald
1223  * VdW interaction:            LennardJones
1224  * Geometry:                   Water3-Water3
1225  * Calculate force/pot:        Force
1226  */
1227 void
1228 nb_kernel_ElecEw_VdwLJ_GeomW3W3_F_avx_128_fma_double
1229                     (t_nblist                    * gmx_restrict       nlist,
1230                      rvec                        * gmx_restrict          xx,
1231                      rvec                        * gmx_restrict          ff,
1232                      t_forcerec                  * gmx_restrict          fr,
1233                      t_mdatoms                   * gmx_restrict     mdatoms,
1234                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1235                      t_nrnb                      * gmx_restrict        nrnb)
1236 {
1237     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1238      * just 0 for non-waters.
1239      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1240      * jnr indices corresponding to data put in the four positions in the SIMD register.
1241      */
1242     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1243     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1244     int              jnrA,jnrB;
1245     int              j_coord_offsetA,j_coord_offsetB;
1246     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1247     real             rcutoff_scalar;
1248     real             *shiftvec,*fshift,*x,*f;
1249     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1250     int              vdwioffset0;
1251     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1252     int              vdwioffset1;
1253     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1254     int              vdwioffset2;
1255     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1256     int              vdwjidx0A,vdwjidx0B;
1257     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1258     int              vdwjidx1A,vdwjidx1B;
1259     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1260     int              vdwjidx2A,vdwjidx2B;
1261     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1262     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1263     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1264     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1265     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1266     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1267     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1268     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1269     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1270     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1271     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1272     real             *charge;
1273     int              nvdwtype;
1274     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1275     int              *vdwtype;
1276     real             *vdwparam;
1277     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1278     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1279     __m128i          ewitab;
1280     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1281     real             *ewtab;
1282     __m128d          dummy_mask,cutoff_mask;
1283     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1284     __m128d          one     = _mm_set1_pd(1.0);
1285     __m128d          two     = _mm_set1_pd(2.0);
1286     x                = xx[0];
1287     f                = ff[0];
1288
1289     nri              = nlist->nri;
1290     iinr             = nlist->iinr;
1291     jindex           = nlist->jindex;
1292     jjnr             = nlist->jjnr;
1293     shiftidx         = nlist->shift;
1294     gid              = nlist->gid;
1295     shiftvec         = fr->shift_vec[0];
1296     fshift           = fr->fshift[0];
1297     facel            = _mm_set1_pd(fr->epsfac);
1298     charge           = mdatoms->chargeA;
1299     nvdwtype         = fr->ntype;
1300     vdwparam         = fr->nbfp;
1301     vdwtype          = mdatoms->typeA;
1302
1303     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1304     ewtab            = fr->ic->tabq_coul_F;
1305     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1306     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1307
1308     /* Setup water-specific parameters */
1309     inr              = nlist->iinr[0];
1310     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1311     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1312     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1313     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1314
1315     jq0              = _mm_set1_pd(charge[inr+0]);
1316     jq1              = _mm_set1_pd(charge[inr+1]);
1317     jq2              = _mm_set1_pd(charge[inr+2]);
1318     vdwjidx0A        = 2*vdwtype[inr+0];
1319     qq00             = _mm_mul_pd(iq0,jq0);
1320     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1321     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1322     qq01             = _mm_mul_pd(iq0,jq1);
1323     qq02             = _mm_mul_pd(iq0,jq2);
1324     qq10             = _mm_mul_pd(iq1,jq0);
1325     qq11             = _mm_mul_pd(iq1,jq1);
1326     qq12             = _mm_mul_pd(iq1,jq2);
1327     qq20             = _mm_mul_pd(iq2,jq0);
1328     qq21             = _mm_mul_pd(iq2,jq1);
1329     qq22             = _mm_mul_pd(iq2,jq2);
1330
1331     /* Avoid stupid compiler warnings */
1332     jnrA = jnrB = 0;
1333     j_coord_offsetA = 0;
1334     j_coord_offsetB = 0;
1335
1336     outeriter        = 0;
1337     inneriter        = 0;
1338
1339     /* Start outer loop over neighborlists */
1340     for(iidx=0; iidx<nri; iidx++)
1341     {
1342         /* Load shift vector for this list */
1343         i_shift_offset   = DIM*shiftidx[iidx];
1344
1345         /* Load limits for loop over neighbors */
1346         j_index_start    = jindex[iidx];
1347         j_index_end      = jindex[iidx+1];
1348
1349         /* Get outer coordinate index */
1350         inr              = iinr[iidx];
1351         i_coord_offset   = DIM*inr;
1352
1353         /* Load i particle coords and add shift vector */
1354         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1355                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1356
1357         fix0             = _mm_setzero_pd();
1358         fiy0             = _mm_setzero_pd();
1359         fiz0             = _mm_setzero_pd();
1360         fix1             = _mm_setzero_pd();
1361         fiy1             = _mm_setzero_pd();
1362         fiz1             = _mm_setzero_pd();
1363         fix2             = _mm_setzero_pd();
1364         fiy2             = _mm_setzero_pd();
1365         fiz2             = _mm_setzero_pd();
1366
1367         /* Start inner kernel loop */
1368         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1369         {
1370
1371             /* Get j neighbor index, and coordinate index */
1372             jnrA             = jjnr[jidx];
1373             jnrB             = jjnr[jidx+1];
1374             j_coord_offsetA  = DIM*jnrA;
1375             j_coord_offsetB  = DIM*jnrB;
1376
1377             /* load j atom coordinates */
1378             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1379                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1380
1381             /* Calculate displacement vector */
1382             dx00             = _mm_sub_pd(ix0,jx0);
1383             dy00             = _mm_sub_pd(iy0,jy0);
1384             dz00             = _mm_sub_pd(iz0,jz0);
1385             dx01             = _mm_sub_pd(ix0,jx1);
1386             dy01             = _mm_sub_pd(iy0,jy1);
1387             dz01             = _mm_sub_pd(iz0,jz1);
1388             dx02             = _mm_sub_pd(ix0,jx2);
1389             dy02             = _mm_sub_pd(iy0,jy2);
1390             dz02             = _mm_sub_pd(iz0,jz2);
1391             dx10             = _mm_sub_pd(ix1,jx0);
1392             dy10             = _mm_sub_pd(iy1,jy0);
1393             dz10             = _mm_sub_pd(iz1,jz0);
1394             dx11             = _mm_sub_pd(ix1,jx1);
1395             dy11             = _mm_sub_pd(iy1,jy1);
1396             dz11             = _mm_sub_pd(iz1,jz1);
1397             dx12             = _mm_sub_pd(ix1,jx2);
1398             dy12             = _mm_sub_pd(iy1,jy2);
1399             dz12             = _mm_sub_pd(iz1,jz2);
1400             dx20             = _mm_sub_pd(ix2,jx0);
1401             dy20             = _mm_sub_pd(iy2,jy0);
1402             dz20             = _mm_sub_pd(iz2,jz0);
1403             dx21             = _mm_sub_pd(ix2,jx1);
1404             dy21             = _mm_sub_pd(iy2,jy1);
1405             dz21             = _mm_sub_pd(iz2,jz1);
1406             dx22             = _mm_sub_pd(ix2,jx2);
1407             dy22             = _mm_sub_pd(iy2,jy2);
1408             dz22             = _mm_sub_pd(iz2,jz2);
1409
1410             /* Calculate squared distance and things based on it */
1411             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1412             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1413             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1414             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1415             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1416             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1417             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1418             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1419             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1420
1421             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1422             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1423             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1424             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1425             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1426             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1427             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1428             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1429             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1430
1431             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1432             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1433             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1434             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1435             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1436             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1437             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1438             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1439             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1440
1441             fjx0             = _mm_setzero_pd();
1442             fjy0             = _mm_setzero_pd();
1443             fjz0             = _mm_setzero_pd();
1444             fjx1             = _mm_setzero_pd();
1445             fjy1             = _mm_setzero_pd();
1446             fjz1             = _mm_setzero_pd();
1447             fjx2             = _mm_setzero_pd();
1448             fjy2             = _mm_setzero_pd();
1449             fjz2             = _mm_setzero_pd();
1450
1451             /**************************
1452              * CALCULATE INTERACTIONS *
1453              **************************/
1454
1455             r00              = _mm_mul_pd(rsq00,rinv00);
1456
1457             /* EWALD ELECTROSTATICS */
1458
1459             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1460             ewrt             = _mm_mul_pd(r00,ewtabscale);
1461             ewitab           = _mm_cvttpd_epi32(ewrt);
1462 #ifdef __XOP__
1463             eweps            = _mm_frcz_pd(ewrt);
1464 #else
1465             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1466 #endif
1467             twoeweps         = _mm_add_pd(eweps,eweps);
1468             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1469                                          &ewtabF,&ewtabFn);
1470             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1471             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1472
1473             /* LENNARD-JONES DISPERSION/REPULSION */
1474
1475             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1476             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1477
1478             fscal            = _mm_add_pd(felec,fvdw);
1479
1480             /* Update vectorial force */
1481             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1482             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1483             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1484
1485             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1486             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1487             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1488
1489             /**************************
1490              * CALCULATE INTERACTIONS *
1491              **************************/
1492
1493             r01              = _mm_mul_pd(rsq01,rinv01);
1494
1495             /* EWALD ELECTROSTATICS */
1496
1497             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1498             ewrt             = _mm_mul_pd(r01,ewtabscale);
1499             ewitab           = _mm_cvttpd_epi32(ewrt);
1500 #ifdef __XOP__
1501             eweps            = _mm_frcz_pd(ewrt);
1502 #else
1503             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1504 #endif
1505             twoeweps         = _mm_add_pd(eweps,eweps);
1506             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1507                                          &ewtabF,&ewtabFn);
1508             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1509             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1510
1511             fscal            = felec;
1512
1513             /* Update vectorial force */
1514             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1515             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1516             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1517
1518             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1519             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1520             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1521
1522             /**************************
1523              * CALCULATE INTERACTIONS *
1524              **************************/
1525
1526             r02              = _mm_mul_pd(rsq02,rinv02);
1527
1528             /* EWALD ELECTROSTATICS */
1529
1530             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1531             ewrt             = _mm_mul_pd(r02,ewtabscale);
1532             ewitab           = _mm_cvttpd_epi32(ewrt);
1533 #ifdef __XOP__
1534             eweps            = _mm_frcz_pd(ewrt);
1535 #else
1536             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1537 #endif
1538             twoeweps         = _mm_add_pd(eweps,eweps);
1539             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1540                                          &ewtabF,&ewtabFn);
1541             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1542             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1543
1544             fscal            = felec;
1545
1546             /* Update vectorial force */
1547             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1548             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1549             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1550
1551             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1552             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1553             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1554
1555             /**************************
1556              * CALCULATE INTERACTIONS *
1557              **************************/
1558
1559             r10              = _mm_mul_pd(rsq10,rinv10);
1560
1561             /* EWALD ELECTROSTATICS */
1562
1563             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1564             ewrt             = _mm_mul_pd(r10,ewtabscale);
1565             ewitab           = _mm_cvttpd_epi32(ewrt);
1566 #ifdef __XOP__
1567             eweps            = _mm_frcz_pd(ewrt);
1568 #else
1569             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1570 #endif
1571             twoeweps         = _mm_add_pd(eweps,eweps);
1572             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1573                                          &ewtabF,&ewtabFn);
1574             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1575             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1576
1577             fscal            = felec;
1578
1579             /* Update vectorial force */
1580             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1581             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1582             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1583
1584             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1585             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1586             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1587
1588             /**************************
1589              * CALCULATE INTERACTIONS *
1590              **************************/
1591
1592             r11              = _mm_mul_pd(rsq11,rinv11);
1593
1594             /* EWALD ELECTROSTATICS */
1595
1596             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1597             ewrt             = _mm_mul_pd(r11,ewtabscale);
1598             ewitab           = _mm_cvttpd_epi32(ewrt);
1599 #ifdef __XOP__
1600             eweps            = _mm_frcz_pd(ewrt);
1601 #else
1602             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1603 #endif
1604             twoeweps         = _mm_add_pd(eweps,eweps);
1605             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1606                                          &ewtabF,&ewtabFn);
1607             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1608             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1609
1610             fscal            = felec;
1611
1612             /* Update vectorial force */
1613             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1614             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1615             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1616
1617             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1618             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1619             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1620
1621             /**************************
1622              * CALCULATE INTERACTIONS *
1623              **************************/
1624
1625             r12              = _mm_mul_pd(rsq12,rinv12);
1626
1627             /* EWALD ELECTROSTATICS */
1628
1629             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1630             ewrt             = _mm_mul_pd(r12,ewtabscale);
1631             ewitab           = _mm_cvttpd_epi32(ewrt);
1632 #ifdef __XOP__
1633             eweps            = _mm_frcz_pd(ewrt);
1634 #else
1635             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1636 #endif
1637             twoeweps         = _mm_add_pd(eweps,eweps);
1638             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1639                                          &ewtabF,&ewtabFn);
1640             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1641             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1642
1643             fscal            = felec;
1644
1645             /* Update vectorial force */
1646             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1647             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1648             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1649
1650             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1651             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1652             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1653
1654             /**************************
1655              * CALCULATE INTERACTIONS *
1656              **************************/
1657
1658             r20              = _mm_mul_pd(rsq20,rinv20);
1659
1660             /* EWALD ELECTROSTATICS */
1661
1662             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1663             ewrt             = _mm_mul_pd(r20,ewtabscale);
1664             ewitab           = _mm_cvttpd_epi32(ewrt);
1665 #ifdef __XOP__
1666             eweps            = _mm_frcz_pd(ewrt);
1667 #else
1668             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1669 #endif
1670             twoeweps         = _mm_add_pd(eweps,eweps);
1671             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1672                                          &ewtabF,&ewtabFn);
1673             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1674             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1675
1676             fscal            = felec;
1677
1678             /* Update vectorial force */
1679             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1680             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1681             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1682
1683             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1684             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1685             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1686
1687             /**************************
1688              * CALCULATE INTERACTIONS *
1689              **************************/
1690
1691             r21              = _mm_mul_pd(rsq21,rinv21);
1692
1693             /* EWALD ELECTROSTATICS */
1694
1695             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1696             ewrt             = _mm_mul_pd(r21,ewtabscale);
1697             ewitab           = _mm_cvttpd_epi32(ewrt);
1698 #ifdef __XOP__
1699             eweps            = _mm_frcz_pd(ewrt);
1700 #else
1701             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1702 #endif
1703             twoeweps         = _mm_add_pd(eweps,eweps);
1704             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1705                                          &ewtabF,&ewtabFn);
1706             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1707             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1708
1709             fscal            = felec;
1710
1711             /* Update vectorial force */
1712             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1713             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1714             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1715
1716             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1717             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1718             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1719
1720             /**************************
1721              * CALCULATE INTERACTIONS *
1722              **************************/
1723
1724             r22              = _mm_mul_pd(rsq22,rinv22);
1725
1726             /* EWALD ELECTROSTATICS */
1727
1728             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1729             ewrt             = _mm_mul_pd(r22,ewtabscale);
1730             ewitab           = _mm_cvttpd_epi32(ewrt);
1731 #ifdef __XOP__
1732             eweps            = _mm_frcz_pd(ewrt);
1733 #else
1734             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1735 #endif
1736             twoeweps         = _mm_add_pd(eweps,eweps);
1737             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1738                                          &ewtabF,&ewtabFn);
1739             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1740             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1741
1742             fscal            = felec;
1743
1744             /* Update vectorial force */
1745             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1746             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1747             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1748
1749             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1750             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1751             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1752
1753             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1754
1755             /* Inner loop uses 358 flops */
1756         }
1757
1758         if(jidx<j_index_end)
1759         {
1760
1761             jnrA             = jjnr[jidx];
1762             j_coord_offsetA  = DIM*jnrA;
1763
1764             /* load j atom coordinates */
1765             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1766                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1767
1768             /* Calculate displacement vector */
1769             dx00             = _mm_sub_pd(ix0,jx0);
1770             dy00             = _mm_sub_pd(iy0,jy0);
1771             dz00             = _mm_sub_pd(iz0,jz0);
1772             dx01             = _mm_sub_pd(ix0,jx1);
1773             dy01             = _mm_sub_pd(iy0,jy1);
1774             dz01             = _mm_sub_pd(iz0,jz1);
1775             dx02             = _mm_sub_pd(ix0,jx2);
1776             dy02             = _mm_sub_pd(iy0,jy2);
1777             dz02             = _mm_sub_pd(iz0,jz2);
1778             dx10             = _mm_sub_pd(ix1,jx0);
1779             dy10             = _mm_sub_pd(iy1,jy0);
1780             dz10             = _mm_sub_pd(iz1,jz0);
1781             dx11             = _mm_sub_pd(ix1,jx1);
1782             dy11             = _mm_sub_pd(iy1,jy1);
1783             dz11             = _mm_sub_pd(iz1,jz1);
1784             dx12             = _mm_sub_pd(ix1,jx2);
1785             dy12             = _mm_sub_pd(iy1,jy2);
1786             dz12             = _mm_sub_pd(iz1,jz2);
1787             dx20             = _mm_sub_pd(ix2,jx0);
1788             dy20             = _mm_sub_pd(iy2,jy0);
1789             dz20             = _mm_sub_pd(iz2,jz0);
1790             dx21             = _mm_sub_pd(ix2,jx1);
1791             dy21             = _mm_sub_pd(iy2,jy1);
1792             dz21             = _mm_sub_pd(iz2,jz1);
1793             dx22             = _mm_sub_pd(ix2,jx2);
1794             dy22             = _mm_sub_pd(iy2,jy2);
1795             dz22             = _mm_sub_pd(iz2,jz2);
1796
1797             /* Calculate squared distance and things based on it */
1798             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1799             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1800             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1801             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1802             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1803             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1804             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1805             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1806             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1807
1808             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1809             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1810             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1811             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1812             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1813             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1814             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1815             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1816             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1817
1818             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1819             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1820             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1821             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1822             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1823             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1824             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1825             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1826             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1827
1828             fjx0             = _mm_setzero_pd();
1829             fjy0             = _mm_setzero_pd();
1830             fjz0             = _mm_setzero_pd();
1831             fjx1             = _mm_setzero_pd();
1832             fjy1             = _mm_setzero_pd();
1833             fjz1             = _mm_setzero_pd();
1834             fjx2             = _mm_setzero_pd();
1835             fjy2             = _mm_setzero_pd();
1836             fjz2             = _mm_setzero_pd();
1837
1838             /**************************
1839              * CALCULATE INTERACTIONS *
1840              **************************/
1841
1842             r00              = _mm_mul_pd(rsq00,rinv00);
1843
1844             /* EWALD ELECTROSTATICS */
1845
1846             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1847             ewrt             = _mm_mul_pd(r00,ewtabscale);
1848             ewitab           = _mm_cvttpd_epi32(ewrt);
1849 #ifdef __XOP__
1850             eweps            = _mm_frcz_pd(ewrt);
1851 #else
1852             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1853 #endif
1854             twoeweps         = _mm_add_pd(eweps,eweps);
1855             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1856             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1857             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1858
1859             /* LENNARD-JONES DISPERSION/REPULSION */
1860
1861             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1862             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1863
1864             fscal            = _mm_add_pd(felec,fvdw);
1865
1866             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1867
1868             /* Update vectorial force */
1869             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1870             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1871             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1872
1873             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1874             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1875             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1876
1877             /**************************
1878              * CALCULATE INTERACTIONS *
1879              **************************/
1880
1881             r01              = _mm_mul_pd(rsq01,rinv01);
1882
1883             /* EWALD ELECTROSTATICS */
1884
1885             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1886             ewrt             = _mm_mul_pd(r01,ewtabscale);
1887             ewitab           = _mm_cvttpd_epi32(ewrt);
1888 #ifdef __XOP__
1889             eweps            = _mm_frcz_pd(ewrt);
1890 #else
1891             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1892 #endif
1893             twoeweps         = _mm_add_pd(eweps,eweps);
1894             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1895             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1896             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1897
1898             fscal            = felec;
1899
1900             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1901
1902             /* Update vectorial force */
1903             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1904             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1905             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1906
1907             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1908             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1909             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1910
1911             /**************************
1912              * CALCULATE INTERACTIONS *
1913              **************************/
1914
1915             r02              = _mm_mul_pd(rsq02,rinv02);
1916
1917             /* EWALD ELECTROSTATICS */
1918
1919             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1920             ewrt             = _mm_mul_pd(r02,ewtabscale);
1921             ewitab           = _mm_cvttpd_epi32(ewrt);
1922 #ifdef __XOP__
1923             eweps            = _mm_frcz_pd(ewrt);
1924 #else
1925             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1926 #endif
1927             twoeweps         = _mm_add_pd(eweps,eweps);
1928             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1929             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1930             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1931
1932             fscal            = felec;
1933
1934             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1935
1936             /* Update vectorial force */
1937             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1938             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1939             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1940
1941             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1942             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1943             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1944
1945             /**************************
1946              * CALCULATE INTERACTIONS *
1947              **************************/
1948
1949             r10              = _mm_mul_pd(rsq10,rinv10);
1950
1951             /* EWALD ELECTROSTATICS */
1952
1953             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1954             ewrt             = _mm_mul_pd(r10,ewtabscale);
1955             ewitab           = _mm_cvttpd_epi32(ewrt);
1956 #ifdef __XOP__
1957             eweps            = _mm_frcz_pd(ewrt);
1958 #else
1959             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1960 #endif
1961             twoeweps         = _mm_add_pd(eweps,eweps);
1962             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1963             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1964             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1965
1966             fscal            = felec;
1967
1968             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1969
1970             /* Update vectorial force */
1971             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1972             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1973             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1974
1975             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1976             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1977             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1978
1979             /**************************
1980              * CALCULATE INTERACTIONS *
1981              **************************/
1982
1983             r11              = _mm_mul_pd(rsq11,rinv11);
1984
1985             /* EWALD ELECTROSTATICS */
1986
1987             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1988             ewrt             = _mm_mul_pd(r11,ewtabscale);
1989             ewitab           = _mm_cvttpd_epi32(ewrt);
1990 #ifdef __XOP__
1991             eweps            = _mm_frcz_pd(ewrt);
1992 #else
1993             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1994 #endif
1995             twoeweps         = _mm_add_pd(eweps,eweps);
1996             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1997             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1998             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1999
2000             fscal            = felec;
2001
2002             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2003
2004             /* Update vectorial force */
2005             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2006             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2007             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2008
2009             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2010             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2011             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2012
2013             /**************************
2014              * CALCULATE INTERACTIONS *
2015              **************************/
2016
2017             r12              = _mm_mul_pd(rsq12,rinv12);
2018
2019             /* EWALD ELECTROSTATICS */
2020
2021             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2022             ewrt             = _mm_mul_pd(r12,ewtabscale);
2023             ewitab           = _mm_cvttpd_epi32(ewrt);
2024 #ifdef __XOP__
2025             eweps            = _mm_frcz_pd(ewrt);
2026 #else
2027             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2028 #endif
2029             twoeweps         = _mm_add_pd(eweps,eweps);
2030             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2031             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2032             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2033
2034             fscal            = felec;
2035
2036             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2037
2038             /* Update vectorial force */
2039             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2040             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2041             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2042
2043             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2044             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2045             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2046
2047             /**************************
2048              * CALCULATE INTERACTIONS *
2049              **************************/
2050
2051             r20              = _mm_mul_pd(rsq20,rinv20);
2052
2053             /* EWALD ELECTROSTATICS */
2054
2055             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2056             ewrt             = _mm_mul_pd(r20,ewtabscale);
2057             ewitab           = _mm_cvttpd_epi32(ewrt);
2058 #ifdef __XOP__
2059             eweps            = _mm_frcz_pd(ewrt);
2060 #else
2061             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2062 #endif
2063             twoeweps         = _mm_add_pd(eweps,eweps);
2064             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2065             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2066             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2067
2068             fscal            = felec;
2069
2070             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2071
2072             /* Update vectorial force */
2073             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2074             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2075             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2076
2077             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2078             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2079             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
2080
2081             /**************************
2082              * CALCULATE INTERACTIONS *
2083              **************************/
2084
2085             r21              = _mm_mul_pd(rsq21,rinv21);
2086
2087             /* EWALD ELECTROSTATICS */
2088
2089             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2090             ewrt             = _mm_mul_pd(r21,ewtabscale);
2091             ewitab           = _mm_cvttpd_epi32(ewrt);
2092 #ifdef __XOP__
2093             eweps            = _mm_frcz_pd(ewrt);
2094 #else
2095             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2096 #endif
2097             twoeweps         = _mm_add_pd(eweps,eweps);
2098             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2099             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2100             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2101
2102             fscal            = felec;
2103
2104             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2105
2106             /* Update vectorial force */
2107             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2108             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2109             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2110
2111             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2112             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2113             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2114
2115             /**************************
2116              * CALCULATE INTERACTIONS *
2117              **************************/
2118
2119             r22              = _mm_mul_pd(rsq22,rinv22);
2120
2121             /* EWALD ELECTROSTATICS */
2122
2123             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2124             ewrt             = _mm_mul_pd(r22,ewtabscale);
2125             ewitab           = _mm_cvttpd_epi32(ewrt);
2126 #ifdef __XOP__
2127             eweps            = _mm_frcz_pd(ewrt);
2128 #else
2129             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2130 #endif
2131             twoeweps         = _mm_add_pd(eweps,eweps);
2132             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2133             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2134             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2135
2136             fscal            = felec;
2137
2138             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2139
2140             /* Update vectorial force */
2141             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2142             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2143             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2144
2145             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2146             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2147             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2148
2149             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2150
2151             /* Inner loop uses 358 flops */
2152         }
2153
2154         /* End of innermost loop */
2155
2156         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2157                                               f+i_coord_offset,fshift+i_shift_offset);
2158
2159         /* Increment number of inner iterations */
2160         inneriter                  += j_index_end - j_index_start;
2161
2162         /* Outer loop uses 18 flops */
2163     }
2164
2165     /* Increment number of outer iterations */
2166     outeriter        += nri;
2167
2168     /* Update outer/inner flops */
2169
2170     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*358);
2171 }