src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_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_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_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     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          ewitab;
 102     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 103     real             *ewtab;
 104     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 105     real             rswitch_scalar,d_scalar;
 106     __m128d          dummy_mask,cutoff_mask;
 107     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 108     __m128d          one     = _mm_set1_pd(1.0);
 109     __m128d          two     = _mm_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 128     ewtab            = fr->ic->tabq_coul_FDV0;
 129     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 130     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 135     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 136     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 140     rcutoff_scalar   = fr->rcoulomb;
 141     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 142     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 143
 144     rswitch_scalar   = fr->rcoulomb_switch;
 145     rswitch          = _mm_set1_pd(rswitch_scalar);
 146     /* Setup switch parameters */
 147     d_scalar         = rcutoff_scalar-rswitch_scalar;
 148     d                = _mm_set1_pd(d_scalar);
 149     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 150     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 151     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 152     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 153     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 154     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     /* Start outer loop over neighborlists */
 165     for(iidx=0; iidx<nri; iidx++)
 166     {
 167         /* Load shift vector for this list */
 168         i_shift_offset   = DIM*shiftidx[iidx];
 169
 170         /* Load limits for loop over neighbors */
 171         j_index_start    = jindex[iidx];
 172         j_index_end      = jindex[iidx+1];
 173
 174         /* Get outer coordinate index */
 175         inr              = iinr[iidx];
 176         i_coord_offset   = DIM*inr;
 177
 178         /* Load i particle coords and add shift vector */
 179         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 180                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 181
 182         fix0             = _mm_setzero_pd();
 183         fiy0             = _mm_setzero_pd();
 184         fiz0             = _mm_setzero_pd();
 185         fix1             = _mm_setzero_pd();
 186         fiy1             = _mm_setzero_pd();
 187         fiz1             = _mm_setzero_pd();
 188         fix2             = _mm_setzero_pd();
 189         fiy2             = _mm_setzero_pd();
 190         fiz2             = _mm_setzero_pd();
 191
 192         /* Reset potential sums */
 193         velecsum         = _mm_setzero_pd();
 194         vvdwsum          = _mm_setzero_pd();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm_sub_pd(ix0,jx0);
 212             dy00             = _mm_sub_pd(iy0,jy0);
 213             dz00             = _mm_sub_pd(iz0,jz0);
 214             dx10             = _mm_sub_pd(ix1,jx0);
 215             dy10             = _mm_sub_pd(iy1,jy0);
 216             dz10             = _mm_sub_pd(iz1,jz0);
 217             dx20             = _mm_sub_pd(ix2,jx0);
 218             dy20             = _mm_sub_pd(iy2,jy0);
 219             dz20             = _mm_sub_pd(iz2,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 225
 226             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 227             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 228             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 229
 230             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 231             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 232             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238
 239             fjx0             = _mm_setzero_pd();
 240             fjy0             = _mm_setzero_pd();
 241             fjz0             = _mm_setzero_pd();
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             if (gmx_mm_any_lt(rsq00,rcutoff2))
 248             {
 249
 250             r00              = _mm_mul_pd(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             qq00             = _mm_mul_pd(iq0,jq0);
 254             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 255                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 256
 257             /* EWALD ELECTROSTATICS */
 258
 259             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 260             ewrt             = _mm_mul_pd(r00,ewtabscale);
 261             ewitab           = _mm_cvttpd_epi32(ewrt);
 262 #ifdef __XOP__
 263             eweps            = _mm_frcz_pd(ewrt);
 264 #else
 265             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 266 #endif
 267             twoeweps         = _mm_add_pd(eweps,eweps);
 268             ewitab           = _mm_slli_epi32(ewitab,2);
 269             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 270             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 271             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 272             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 273             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 274             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 275             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 276             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 277             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 278             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 279
 280             /* LENNARD-JONES DISPERSION/REPULSION */
 281
 282             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 283             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 284             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 285             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 286             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 287
 288             d                = _mm_sub_pd(r00,rswitch);
 289             d                = _mm_max_pd(d,_mm_setzero_pd());
 290             d2               = _mm_mul_pd(d,d);
 291             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 292
 293             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 294
 295             /* Evaluate switch function */
 296             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 297             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 298             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 299             velec            = _mm_mul_pd(velec,sw);
 300             vvdw             = _mm_mul_pd(vvdw,sw);
 301             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velec            = _mm_and_pd(velec,cutoff_mask);
 305             velecsum         = _mm_add_pd(velecsum,velec);
 306             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 307             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 308
 309             fscal            = _mm_add_pd(felec,fvdw);
 310
 311             fscal            = _mm_and_pd(fscal,cutoff_mask);
 312
 313             /* Update vectorial force */
 314             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 315             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 316             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 317
 318             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 319             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 320             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 321
 322             }
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             if (gmx_mm_any_lt(rsq10,rcutoff2))
 329             {
 330
 331             r10              = _mm_mul_pd(rsq10,rinv10);
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             qq10             = _mm_mul_pd(iq1,jq0);
 335
 336             /* EWALD ELECTROSTATICS */
 337
 338             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 339             ewrt             = _mm_mul_pd(r10,ewtabscale);
 340             ewitab           = _mm_cvttpd_epi32(ewrt);
 341 #ifdef __XOP__
 342             eweps            = _mm_frcz_pd(ewrt);
 343 #else
 344             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 345 #endif
 346             twoeweps         = _mm_add_pd(eweps,eweps);
 347             ewitab           = _mm_slli_epi32(ewitab,2);
 348             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 349             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 350             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 351             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 352             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 353             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 354             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 355             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 356             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 357             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 358
 359             d                = _mm_sub_pd(r10,rswitch);
 360             d                = _mm_max_pd(d,_mm_setzero_pd());
 361             d2               = _mm_mul_pd(d,d);
 362             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 363
 364             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 365
 366             /* Evaluate switch function */
 367             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 368             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 369             velec            = _mm_mul_pd(velec,sw);
 370             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velec            = _mm_and_pd(velec,cutoff_mask);
 374             velecsum         = _mm_add_pd(velecsum,velec);
 375
 376             fscal            = felec;
 377
 378             fscal            = _mm_and_pd(fscal,cutoff_mask);
 379
 380             /* Update vectorial force */
 381             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 382             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 383             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 384
 385             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 386             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 387             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 388
 389             }
 390
 391             /**************************
 392              * CALCULATE INTERACTIONS *
 393              **************************/
 394
 395             if (gmx_mm_any_lt(rsq20,rcutoff2))
 396             {
 397
 398             r20              = _mm_mul_pd(rsq20,rinv20);
 399
 400             /* Compute parameters for interactions between i and j atoms */
 401             qq20             = _mm_mul_pd(iq2,jq0);
 402
 403             /* EWALD ELECTROSTATICS */
 404
 405             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 406             ewrt             = _mm_mul_pd(r20,ewtabscale);
 407             ewitab           = _mm_cvttpd_epi32(ewrt);
 408 #ifdef __XOP__
 409             eweps            = _mm_frcz_pd(ewrt);
 410 #else
 411             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 412 #endif
 413             twoeweps         = _mm_add_pd(eweps,eweps);
 414             ewitab           = _mm_slli_epi32(ewitab,2);
 415             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 416             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 417             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 418             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 419             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 420             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 421             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 422             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 423             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 424             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 425
 426             d                = _mm_sub_pd(r20,rswitch);
 427             d                = _mm_max_pd(d,_mm_setzero_pd());
 428             d2               = _mm_mul_pd(d,d);
 429             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 430
 431             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 432
 433             /* Evaluate switch function */
 434             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 435             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 436             velec            = _mm_mul_pd(velec,sw);
 437             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 438
 439             /* Update potential sum for this i atom from the interaction with this j atom. */
 440             velec            = _mm_and_pd(velec,cutoff_mask);
 441             velecsum         = _mm_add_pd(velecsum,velec);
 442
 443             fscal            = felec;
 444
 445             fscal            = _mm_and_pd(fscal,cutoff_mask);
 446
 447             /* Update vectorial force */
 448             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 449             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 450             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 451
 452             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 453             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 454             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 455
 456             }
 457
 458             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 459
 460             /* Inner loop uses 225 flops */
 461         }
 462
 463         if(jidx<j_index_end)
 464         {
 465
 466             jnrA             = jjnr[jidx];
 467             j_coord_offsetA  = DIM*jnrA;
 468
 469             /* load j atom coordinates */
 470             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 471                                               &jx0,&jy0,&jz0);
 472
 473             /* Calculate displacement vector */
 474             dx00             = _mm_sub_pd(ix0,jx0);
 475             dy00             = _mm_sub_pd(iy0,jy0);
 476             dz00             = _mm_sub_pd(iz0,jz0);
 477             dx10             = _mm_sub_pd(ix1,jx0);
 478             dy10             = _mm_sub_pd(iy1,jy0);
 479             dz10             = _mm_sub_pd(iz1,jz0);
 480             dx20             = _mm_sub_pd(ix2,jx0);
 481             dy20             = _mm_sub_pd(iy2,jy0);
 482             dz20             = _mm_sub_pd(iz2,jz0);
 483
 484             /* Calculate squared distance and things based on it */
 485             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 486             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 487             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 488
 489             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 490             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 491             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 492
 493             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 494             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 495             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 496
 497             /* Load parameters for j particles */
 498             jq0              = _mm_load_sd(charge+jnrA+0);
 499             vdwjidx0A        = 2*vdwtype[jnrA+0];
 500
 501             fjx0             = _mm_setzero_pd();
 502             fjy0             = _mm_setzero_pd();
 503             fjz0             = _mm_setzero_pd();
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             if (gmx_mm_any_lt(rsq00,rcutoff2))
 510             {
 511
 512             r00              = _mm_mul_pd(rsq00,rinv00);
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq00             = _mm_mul_pd(iq0,jq0);
 516             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 517
 518             /* EWALD ELECTROSTATICS */
 519
 520             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 521             ewrt             = _mm_mul_pd(r00,ewtabscale);
 522             ewitab           = _mm_cvttpd_epi32(ewrt);
 523 #ifdef __XOP__
 524             eweps            = _mm_frcz_pd(ewrt);
 525 #else
 526             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 527 #endif
 528             twoeweps         = _mm_add_pd(eweps,eweps);
 529             ewitab           = _mm_slli_epi32(ewitab,2);
 530             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 531             ewtabD           = _mm_setzero_pd();
 532             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 533             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 534             ewtabFn          = _mm_setzero_pd();
 535             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 536             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 537             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 538             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 539             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 540
 541             /* LENNARD-JONES DISPERSION/REPULSION */
 542
 543             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 544             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 545             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 546             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 547             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 548
 549             d                = _mm_sub_pd(r00,rswitch);
 550             d                = _mm_max_pd(d,_mm_setzero_pd());
 551             d2               = _mm_mul_pd(d,d);
 552             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 553
 554             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 555
 556             /* Evaluate switch function */
 557             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 558             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 559             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 560             velec            = _mm_mul_pd(velec,sw);
 561             vvdw             = _mm_mul_pd(vvdw,sw);
 562             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             velec            = _mm_and_pd(velec,cutoff_mask);
 566             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 567             velecsum         = _mm_add_pd(velecsum,velec);
 568             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 569             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 570             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 571
 572             fscal            = _mm_add_pd(felec,fvdw);
 573
 574             fscal            = _mm_and_pd(fscal,cutoff_mask);
 575
 576             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 577
 578             /* Update vectorial force */
 579             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 580             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 581             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 582
 583             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 584             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 585             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 586
 587             }
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             if (gmx_mm_any_lt(rsq10,rcutoff2))
 594             {
 595
 596             r10              = _mm_mul_pd(rsq10,rinv10);
 597
 598             /* Compute parameters for interactions between i and j atoms */
 599             qq10             = _mm_mul_pd(iq1,jq0);
 600
 601             /* EWALD ELECTROSTATICS */
 602
 603             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 604             ewrt             = _mm_mul_pd(r10,ewtabscale);
 605             ewitab           = _mm_cvttpd_epi32(ewrt);
 606 #ifdef __XOP__
 607             eweps            = _mm_frcz_pd(ewrt);
 608 #else
 609             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 610 #endif
 611             twoeweps         = _mm_add_pd(eweps,eweps);
 612             ewitab           = _mm_slli_epi32(ewitab,2);
 613             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 614             ewtabD           = _mm_setzero_pd();
 615             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 616             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 617             ewtabFn          = _mm_setzero_pd();
 618             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 619             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 620             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 621             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 622             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 623
 624             d                = _mm_sub_pd(r10,rswitch);
 625             d                = _mm_max_pd(d,_mm_setzero_pd());
 626             d2               = _mm_mul_pd(d,d);
 627             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 628
 629             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 630
 631             /* Evaluate switch function */
 632             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 633             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 634             velec            = _mm_mul_pd(velec,sw);
 635             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 636
 637             /* Update potential sum for this i atom from the interaction with this j atom. */
 638             velec            = _mm_and_pd(velec,cutoff_mask);
 639             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 640             velecsum         = _mm_add_pd(velecsum,velec);
 641
 642             fscal            = felec;
 643
 644             fscal            = _mm_and_pd(fscal,cutoff_mask);
 645
 646             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 647
 648             /* Update vectorial force */
 649             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 650             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 651             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 652
 653             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 654             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 655             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 656
 657             }
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             if (gmx_mm_any_lt(rsq20,rcutoff2))
 664             {
 665
 666             r20              = _mm_mul_pd(rsq20,rinv20);
 667
 668             /* Compute parameters for interactions between i and j atoms */
 669             qq20             = _mm_mul_pd(iq2,jq0);
 670
 671             /* EWALD ELECTROSTATICS */
 672
 673             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 674             ewrt             = _mm_mul_pd(r20,ewtabscale);
 675             ewitab           = _mm_cvttpd_epi32(ewrt);
 676 #ifdef __XOP__
 677             eweps            = _mm_frcz_pd(ewrt);
 678 #else
 679             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 680 #endif
 681             twoeweps         = _mm_add_pd(eweps,eweps);
 682             ewitab           = _mm_slli_epi32(ewitab,2);
 683             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 684             ewtabD           = _mm_setzero_pd();
 685             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 686             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 687             ewtabFn          = _mm_setzero_pd();
 688             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 689             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 690             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 691             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 692             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 693
 694             d                = _mm_sub_pd(r20,rswitch);
 695             d                = _mm_max_pd(d,_mm_setzero_pd());
 696             d2               = _mm_mul_pd(d,d);
 697             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 698
 699             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 700
 701             /* Evaluate switch function */
 702             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 703             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 704             velec            = _mm_mul_pd(velec,sw);
 705             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 706
 707             /* Update potential sum for this i atom from the interaction with this j atom. */
 708             velec            = _mm_and_pd(velec,cutoff_mask);
 709             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 710             velecsum         = _mm_add_pd(velecsum,velec);
 711
 712             fscal            = felec;
 713
 714             fscal            = _mm_and_pd(fscal,cutoff_mask);
 715
 716             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 717
 718             /* Update vectorial force */
 719             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 720             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 721             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 722
 723             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 724             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 725             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 726
 727             }
 728
 729             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 730
 731             /* Inner loop uses 225 flops */
 732         }
 733
 734         /* End of innermost loop */
 735
 736         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 737                                               f+i_coord_offset,fshift+i_shift_offset);
 738
 739         ggid                        = gid[iidx];
 740         /* Update potential energies */
 741         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 742         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 743
 744         /* Increment number of inner iterations */
 745         inneriter                  += j_index_end - j_index_start;
 746
 747         /* Outer loop uses 20 flops */
 748     }
 749
 750     /* Increment number of outer iterations */
 751     outeriter        += nri;
 752
 753     /* Update outer/inner flops */
 754
 755     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*225);
 756 }
 757 /*
 758  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 759  * Electrostatics interaction: Ewald
 760  * VdW interaction:            LennardJones
 761  * Geometry:                   Water3-Particle
 762  * Calculate force/pot:        Force
 763  */
 764 void
 765 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 766                     (t_nblist                    * gmx_restrict       nlist,
 767                      rvec                        * gmx_restrict          xx,
 768                      rvec                        * gmx_restrict          ff,
 769                      t_forcerec                  * gmx_restrict          fr,
 770                      t_mdatoms                   * gmx_restrict     mdatoms,
 771                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 772                      t_nrnb                      * gmx_restrict        nrnb)
 773 {
 774     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 775      * just 0 for non-waters.
 776      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 777      * jnr indices corresponding to data put in the four positions in the SIMD register.
 778      */
 779     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 780     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 781     int              jnrA,jnrB;
 782     int              j_coord_offsetA,j_coord_offsetB;
 783     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 784     real             rcutoff_scalar;
 785     real             *shiftvec,*fshift,*x,*f;
 786     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 787     int              vdwioffset0;
 788     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 789     int              vdwioffset1;
 790     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 791     int              vdwioffset2;
 792     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 793     int              vdwjidx0A,vdwjidx0B;
 794     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 795     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 796     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 797     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 798     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 799     real             *charge;
 800     int              nvdwtype;
 801     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 802     int              *vdwtype;
 803     real             *vdwparam;
 804     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 805     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 806     __m128i          ewitab;
 807     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 808     real             *ewtab;
 809     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 810     real             rswitch_scalar,d_scalar;
 811     __m128d          dummy_mask,cutoff_mask;
 812     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 813     __m128d          one     = _mm_set1_pd(1.0);
 814     __m128d          two     = _mm_set1_pd(2.0);
 815     x                = xx[0];
 816     f                = ff[0];
 817
 818     nri              = nlist->nri;
 819     iinr             = nlist->iinr;
 820     jindex           = nlist->jindex;
 821     jjnr             = nlist->jjnr;
 822     shiftidx         = nlist->shift;
 823     gid              = nlist->gid;
 824     shiftvec         = fr->shift_vec[0];
 825     fshift           = fr->fshift[0];
 826     facel            = _mm_set1_pd(fr->epsfac);
 827     charge           = mdatoms->chargeA;
 828     nvdwtype         = fr->ntype;
 829     vdwparam         = fr->nbfp;
 830     vdwtype          = mdatoms->typeA;
 831
 832     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 833     ewtab            = fr->ic->tabq_coul_FDV0;
 834     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 835     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 836
 837     /* Setup water-specific parameters */
 838     inr              = nlist->iinr[0];
 839     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 840     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 841     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 842     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 843
 844     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 845     rcutoff_scalar   = fr->rcoulomb;
 846     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 847     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 848
 849     rswitch_scalar   = fr->rcoulomb_switch;
 850     rswitch          = _mm_set1_pd(rswitch_scalar);
 851     /* Setup switch parameters */
 852     d_scalar         = rcutoff_scalar-rswitch_scalar;
 853     d                = _mm_set1_pd(d_scalar);
 854     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 855     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 856     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 857     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 858     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 859     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 860
 861     /* Avoid stupid compiler warnings */
 862     jnrA = jnrB = 0;
 863     j_coord_offsetA = 0;
 864     j_coord_offsetB = 0;
 865
 866     outeriter        = 0;
 867     inneriter        = 0;
 868
 869     /* Start outer loop over neighborlists */
 870     for(iidx=0; iidx<nri; iidx++)
 871     {
 872         /* Load shift vector for this list */
 873         i_shift_offset   = DIM*shiftidx[iidx];
 874
 875         /* Load limits for loop over neighbors */
 876         j_index_start    = jindex[iidx];
 877         j_index_end      = jindex[iidx+1];
 878
 879         /* Get outer coordinate index */
 880         inr              = iinr[iidx];
 881         i_coord_offset   = DIM*inr;
 882
 883         /* Load i particle coords and add shift vector */
 884         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 885                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 886
 887         fix0             = _mm_setzero_pd();
 888         fiy0             = _mm_setzero_pd();
 889         fiz0             = _mm_setzero_pd();
 890         fix1             = _mm_setzero_pd();
 891         fiy1             = _mm_setzero_pd();
 892         fiz1             = _mm_setzero_pd();
 893         fix2             = _mm_setzero_pd();
 894         fiy2             = _mm_setzero_pd();
 895         fiz2             = _mm_setzero_pd();
 896
 897         /* Start inner kernel loop */
 898         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 899         {
 900
 901             /* Get j neighbor index, and coordinate index */
 902             jnrA             = jjnr[jidx];
 903             jnrB             = jjnr[jidx+1];
 904             j_coord_offsetA  = DIM*jnrA;
 905             j_coord_offsetB  = DIM*jnrB;
 906
 907             /* load j atom coordinates */
 908             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 909                                               &jx0,&jy0,&jz0);
 910
 911             /* Calculate displacement vector */
 912             dx00             = _mm_sub_pd(ix0,jx0);
 913             dy00             = _mm_sub_pd(iy0,jy0);
 914             dz00             = _mm_sub_pd(iz0,jz0);
 915             dx10             = _mm_sub_pd(ix1,jx0);
 916             dy10             = _mm_sub_pd(iy1,jy0);
 917             dz10             = _mm_sub_pd(iz1,jz0);
 918             dx20             = _mm_sub_pd(ix2,jx0);
 919             dy20             = _mm_sub_pd(iy2,jy0);
 920             dz20             = _mm_sub_pd(iz2,jz0);
 921
 922             /* Calculate squared distance and things based on it */
 923             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 924             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 925             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 926
 927             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 928             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 929             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 930
 931             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 932             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 933             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 934
 935             /* Load parameters for j particles */
 936             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 937             vdwjidx0A        = 2*vdwtype[jnrA+0];
 938             vdwjidx0B        = 2*vdwtype[jnrB+0];
 939
 940             fjx0             = _mm_setzero_pd();
 941             fjy0             = _mm_setzero_pd();
 942             fjz0             = _mm_setzero_pd();
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             if (gmx_mm_any_lt(rsq00,rcutoff2))
 949             {
 950
 951             r00              = _mm_mul_pd(rsq00,rinv00);
 952
 953             /* Compute parameters for interactions between i and j atoms */
 954             qq00             = _mm_mul_pd(iq0,jq0);
 955             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 956                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 957
 958             /* EWALD ELECTROSTATICS */
 959
 960             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 961             ewrt             = _mm_mul_pd(r00,ewtabscale);
 962             ewitab           = _mm_cvttpd_epi32(ewrt);
 963 #ifdef __XOP__
 964             eweps            = _mm_frcz_pd(ewrt);
 965 #else
 966             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 967 #endif
 968             twoeweps         = _mm_add_pd(eweps,eweps);
 969             ewitab           = _mm_slli_epi32(ewitab,2);
 970             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 971             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 972             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 973             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 974             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 975             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 976             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 977             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 978             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 979             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 980
 981             /* LENNARD-JONES DISPERSION/REPULSION */
 982
 983             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 984             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 985             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 986             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 987             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 988
 989             d                = _mm_sub_pd(r00,rswitch);
 990             d                = _mm_max_pd(d,_mm_setzero_pd());
 991             d2               = _mm_mul_pd(d,d);
 992             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 993
 994             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 995
 996             /* Evaluate switch function */
 997             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 998             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 999             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1000             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1001
1002             fscal            = _mm_add_pd(felec,fvdw);
1003
1004             fscal            = _mm_and_pd(fscal,cutoff_mask);
1005
1006             /* Update vectorial force */
1007             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1008             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1009             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1010
1011             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1012             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1013             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1014
1015             }
1016
1017             /**************************
1018              * CALCULATE INTERACTIONS *
1019              **************************/
1020
1021             if (gmx_mm_any_lt(rsq10,rcutoff2))
1022             {
1023
1024             r10              = _mm_mul_pd(rsq10,rinv10);
1025
1026             /* Compute parameters for interactions between i and j atoms */
1027             qq10             = _mm_mul_pd(iq1,jq0);
1028
1029             /* EWALD ELECTROSTATICS */
1030
1031             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1032             ewrt             = _mm_mul_pd(r10,ewtabscale);
1033             ewitab           = _mm_cvttpd_epi32(ewrt);
1034 #ifdef __XOP__
1035             eweps            = _mm_frcz_pd(ewrt);
1036 #else
1037             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1038 #endif
1039             twoeweps         = _mm_add_pd(eweps,eweps);
1040             ewitab           = _mm_slli_epi32(ewitab,2);
1041             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1042             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1043             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1044             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1045             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1046             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1047             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1048             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1049             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1050             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1051
1052             d                = _mm_sub_pd(r10,rswitch);
1053             d                = _mm_max_pd(d,_mm_setzero_pd());
1054             d2               = _mm_mul_pd(d,d);
1055             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1056
1057             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1058
1059             /* Evaluate switch function */
1060             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1061             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1062             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1063
1064             fscal            = felec;
1065
1066             fscal            = _mm_and_pd(fscal,cutoff_mask);
1067
1068             /* Update vectorial force */
1069             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1070             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1071             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1072
1073             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1074             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1075             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1076
1077             }
1078
1079             /**************************
1080              * CALCULATE INTERACTIONS *
1081              **************************/
1082
1083             if (gmx_mm_any_lt(rsq20,rcutoff2))
1084             {
1085
1086             r20              = _mm_mul_pd(rsq20,rinv20);
1087
1088             /* Compute parameters for interactions between i and j atoms */
1089             qq20             = _mm_mul_pd(iq2,jq0);
1090
1091             /* EWALD ELECTROSTATICS */
1092
1093             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1094             ewrt             = _mm_mul_pd(r20,ewtabscale);
1095             ewitab           = _mm_cvttpd_epi32(ewrt);
1096 #ifdef __XOP__
1097             eweps            = _mm_frcz_pd(ewrt);
1098 #else
1099             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1100 #endif
1101             twoeweps         = _mm_add_pd(eweps,eweps);
1102             ewitab           = _mm_slli_epi32(ewitab,2);
1103             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1104             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1105             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1106             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1107             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1108             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1109             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1110             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1111             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1112             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1113
1114             d                = _mm_sub_pd(r20,rswitch);
1115             d                = _mm_max_pd(d,_mm_setzero_pd());
1116             d2               = _mm_mul_pd(d,d);
1117             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1118
1119             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1120
1121             /* Evaluate switch function */
1122             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1123             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1124             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1125
1126             fscal            = felec;
1127
1128             fscal            = _mm_and_pd(fscal,cutoff_mask);
1129
1130             /* Update vectorial force */
1131             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1132             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1133             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1134
1135             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1136             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1137             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1138
1139             }
1140
1141             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1142
1143             /* Inner loop uses 213 flops */
1144         }
1145
1146         if(jidx<j_index_end)
1147         {
1148
1149             jnrA             = jjnr[jidx];
1150             j_coord_offsetA  = DIM*jnrA;
1151
1152             /* load j atom coordinates */
1153             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1154                                               &jx0,&jy0,&jz0);
1155
1156             /* Calculate displacement vector */
1157             dx00             = _mm_sub_pd(ix0,jx0);
1158             dy00             = _mm_sub_pd(iy0,jy0);
1159             dz00             = _mm_sub_pd(iz0,jz0);
1160             dx10             = _mm_sub_pd(ix1,jx0);
1161             dy10             = _mm_sub_pd(iy1,jy0);
1162             dz10             = _mm_sub_pd(iz1,jz0);
1163             dx20             = _mm_sub_pd(ix2,jx0);
1164             dy20             = _mm_sub_pd(iy2,jy0);
1165             dz20             = _mm_sub_pd(iz2,jz0);
1166
1167             /* Calculate squared distance and things based on it */
1168             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1169             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1170             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1171
1172             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1173             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1174             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1175
1176             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1177             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1178             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1179
1180             /* Load parameters for j particles */
1181             jq0              = _mm_load_sd(charge+jnrA+0);
1182             vdwjidx0A        = 2*vdwtype[jnrA+0];
1183
1184             fjx0             = _mm_setzero_pd();
1185             fjy0             = _mm_setzero_pd();
1186             fjz0             = _mm_setzero_pd();
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             if (gmx_mm_any_lt(rsq00,rcutoff2))
1193             {
1194
1195             r00              = _mm_mul_pd(rsq00,rinv00);
1196
1197             /* Compute parameters for interactions between i and j atoms */
1198             qq00             = _mm_mul_pd(iq0,jq0);
1199             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1200
1201             /* EWALD ELECTROSTATICS */
1202
1203             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1204             ewrt             = _mm_mul_pd(r00,ewtabscale);
1205             ewitab           = _mm_cvttpd_epi32(ewrt);
1206 #ifdef __XOP__
1207             eweps            = _mm_frcz_pd(ewrt);
1208 #else
1209             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1210 #endif
1211             twoeweps         = _mm_add_pd(eweps,eweps);
1212             ewitab           = _mm_slli_epi32(ewitab,2);
1213             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1214             ewtabD           = _mm_setzero_pd();
1215             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1216             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1217             ewtabFn          = _mm_setzero_pd();
1218             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1219             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1220             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1221             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
1222             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1223
1224             /* LENNARD-JONES DISPERSION/REPULSION */
1225
1226             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1227             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1228             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1229             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
1230             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1231
1232             d                = _mm_sub_pd(r00,rswitch);
1233             d                = _mm_max_pd(d,_mm_setzero_pd());
1234             d2               = _mm_mul_pd(d,d);
1235             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1236
1237             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1238
1239             /* Evaluate switch function */
1240             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1241             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
1242             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1243             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1244
1245             fscal            = _mm_add_pd(felec,fvdw);
1246
1247             fscal            = _mm_and_pd(fscal,cutoff_mask);
1248
1249             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1250
1251             /* Update vectorial force */
1252             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1253             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1254             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1255
1256             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1257             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1258             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1259
1260             }
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             if (gmx_mm_any_lt(rsq10,rcutoff2))
1267             {
1268
1269             r10              = _mm_mul_pd(rsq10,rinv10);
1270
1271             /* Compute parameters for interactions between i and j atoms */
1272             qq10             = _mm_mul_pd(iq1,jq0);
1273
1274             /* EWALD ELECTROSTATICS */
1275
1276             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1277             ewrt             = _mm_mul_pd(r10,ewtabscale);
1278             ewitab           = _mm_cvttpd_epi32(ewrt);
1279 #ifdef __XOP__
1280             eweps            = _mm_frcz_pd(ewrt);
1281 #else
1282             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1283 #endif
1284             twoeweps         = _mm_add_pd(eweps,eweps);
1285             ewitab           = _mm_slli_epi32(ewitab,2);
1286             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1287             ewtabD           = _mm_setzero_pd();
1288             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1289             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1290             ewtabFn          = _mm_setzero_pd();
1291             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1292             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1293             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1294             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1295             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1296
1297             d                = _mm_sub_pd(r10,rswitch);
1298             d                = _mm_max_pd(d,_mm_setzero_pd());
1299             d2               = _mm_mul_pd(d,d);
1300             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1301
1302             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1303
1304             /* Evaluate switch function */
1305             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1306             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1307             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1308
1309             fscal            = felec;
1310
1311             fscal            = _mm_and_pd(fscal,cutoff_mask);
1312
1313             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1314
1315             /* Update vectorial force */
1316             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1317             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1318             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1319
1320             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1321             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1322             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1323
1324             }
1325
1326             /**************************
1327              * CALCULATE INTERACTIONS *
1328              **************************/
1329
1330             if (gmx_mm_any_lt(rsq20,rcutoff2))
1331             {
1332
1333             r20              = _mm_mul_pd(rsq20,rinv20);
1334
1335             /* Compute parameters for interactions between i and j atoms */
1336             qq20             = _mm_mul_pd(iq2,jq0);
1337
1338             /* EWALD ELECTROSTATICS */
1339
1340             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1341             ewrt             = _mm_mul_pd(r20,ewtabscale);
1342             ewitab           = _mm_cvttpd_epi32(ewrt);
1343 #ifdef __XOP__
1344             eweps            = _mm_frcz_pd(ewrt);
1345 #else
1346             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1347 #endif
1348             twoeweps         = _mm_add_pd(eweps,eweps);
1349             ewitab           = _mm_slli_epi32(ewitab,2);
1350             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1351             ewtabD           = _mm_setzero_pd();
1352             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1353             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1354             ewtabFn          = _mm_setzero_pd();
1355             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1356             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1357             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1358             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1359             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1360
1361             d                = _mm_sub_pd(r20,rswitch);
1362             d                = _mm_max_pd(d,_mm_setzero_pd());
1363             d2               = _mm_mul_pd(d,d);
1364             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1365
1366             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1367
1368             /* Evaluate switch function */
1369             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1370             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1371             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1372
1373             fscal            = felec;
1374
1375             fscal            = _mm_and_pd(fscal,cutoff_mask);
1376
1377             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1378
1379             /* Update vectorial force */
1380             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1381             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1382             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1383
1384             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1385             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1386             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1387
1388             }
1389
1390             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1391
1392             /* Inner loop uses 213 flops */
1393         }
1394
1395         /* End of innermost loop */
1396
1397         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1398                                               f+i_coord_offset,fshift+i_shift_offset);
1399
1400         /* Increment number of inner iterations */
1401         inneriter                  += j_index_end - j_index_start;
1402
1403         /* Outer loop uses 18 flops */
1404     }
1405
1406     /* Increment number of outer iterations */
1407     outeriter        += nri;
1408
1409     /* Update outer/inner flops */
1410
1411     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*213);
1412 }