src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSw_VdwNone_GeomW4P1_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "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_VdwNone_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_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              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128i          ewitab;
  96     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     real             *ewtab;
  98     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  99     real             rswitch_scalar,d_scalar;
 100     __m128d          dummy_mask,cutoff_mask;
 101     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 102     __m128d          one     = _mm_set1_pd(1.0);
 103     __m128d          two     = _mm_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_pd(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117
 118     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 119     ewtab            = fr->ic->tabq_coul_FDV0;
 120     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 121     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 126     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 127     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 128
 129     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 130     rcutoff_scalar   = fr->rcoulomb;
 131     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 132     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 133
 134     rswitch_scalar   = fr->rcoulomb_switch;
 135     rswitch          = _mm_set1_pd(rswitch_scalar);
 136     /* Setup switch parameters */
 137     d_scalar         = rcutoff_scalar-rswitch_scalar;
 138     d                = _mm_set1_pd(d_scalar);
 139     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 140     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 141     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 142     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 143     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 144     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 170                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 171
 172         fix1             = _mm_setzero_pd();
 173         fiy1             = _mm_setzero_pd();
 174         fiz1             = _mm_setzero_pd();
 175         fix2             = _mm_setzero_pd();
 176         fiy2             = _mm_setzero_pd();
 177         fiz2             = _mm_setzero_pd();
 178         fix3             = _mm_setzero_pd();
 179         fiy3             = _mm_setzero_pd();
 180         fiz3             = _mm_setzero_pd();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_pd();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194
 195             /* load j atom coordinates */
 196             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx10             = _mm_sub_pd(ix1,jx0);
 201             dy10             = _mm_sub_pd(iy1,jy0);
 202             dz10             = _mm_sub_pd(iz1,jz0);
 203             dx20             = _mm_sub_pd(ix2,jx0);
 204             dy20             = _mm_sub_pd(iy2,jy0);
 205             dz20             = _mm_sub_pd(iz2,jz0);
 206             dx30             = _mm_sub_pd(ix3,jx0);
 207             dy30             = _mm_sub_pd(iy3,jy0);
 208             dz30             = _mm_sub_pd(iz3,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 213             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 214
 215             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 216             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 217             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 218
 219             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 221             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 225
 226             fjx0             = _mm_setzero_pd();
 227             fjy0             = _mm_setzero_pd();
 228             fjz0             = _mm_setzero_pd();
 229
 230             /**************************
 231              * CALCULATE INTERACTIONS *
 232              **************************/
 233
 234             if (gmx_mm_any_lt(rsq10,rcutoff2))
 235             {
 236
 237             r10              = _mm_mul_pd(rsq10,rinv10);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq10             = _mm_mul_pd(iq1,jq0);
 241
 242             /* EWALD ELECTROSTATICS */
 243
 244             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 245             ewrt             = _mm_mul_pd(r10,ewtabscale);
 246             ewitab           = _mm_cvttpd_epi32(ewrt);
 247 #ifdef __XOP__
 248             eweps            = _mm_frcz_pd(ewrt);
 249 #else
 250             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 251 #endif
 252             twoeweps         = _mm_add_pd(eweps,eweps);
 253             ewitab           = _mm_slli_epi32(ewitab,2);
 254             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 255             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 256             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 257             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 258             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 259             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 260             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 261             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 262             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 263             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 264
 265             d                = _mm_sub_pd(r10,rswitch);
 266             d                = _mm_max_pd(d,_mm_setzero_pd());
 267             d2               = _mm_mul_pd(d,d);
 268             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 269
 270             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 271
 272             /* Evaluate switch function */
 273             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 274             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 275             velec            = _mm_mul_pd(velec,sw);
 276             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velec            = _mm_and_pd(velec,cutoff_mask);
 280             velecsum         = _mm_add_pd(velecsum,velec);
 281
 282             fscal            = felec;
 283
 284             fscal            = _mm_and_pd(fscal,cutoff_mask);
 285
 286             /* Update vectorial force */
 287             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 288             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 289             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 290
 291             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 292             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 293             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 294
 295             }
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             if (gmx_mm_any_lt(rsq20,rcutoff2))
 302             {
 303
 304             r20              = _mm_mul_pd(rsq20,rinv20);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq20             = _mm_mul_pd(iq2,jq0);
 308
 309             /* EWALD ELECTROSTATICS */
 310
 311             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 312             ewrt             = _mm_mul_pd(r20,ewtabscale);
 313             ewitab           = _mm_cvttpd_epi32(ewrt);
 314 #ifdef __XOP__
 315             eweps            = _mm_frcz_pd(ewrt);
 316 #else
 317             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 318 #endif
 319             twoeweps         = _mm_add_pd(eweps,eweps);
 320             ewitab           = _mm_slli_epi32(ewitab,2);
 321             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 322             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 323             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 324             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 325             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 326             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 327             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 328             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 329             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 330             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 331
 332             d                = _mm_sub_pd(r20,rswitch);
 333             d                = _mm_max_pd(d,_mm_setzero_pd());
 334             d2               = _mm_mul_pd(d,d);
 335             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 336
 337             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 338
 339             /* Evaluate switch function */
 340             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 341             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 342             velec            = _mm_mul_pd(velec,sw);
 343             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm_and_pd(velec,cutoff_mask);
 347             velecsum         = _mm_add_pd(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             fscal            = _mm_and_pd(fscal,cutoff_mask);
 352
 353             /* Update vectorial force */
 354             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 355             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 356             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 357
 358             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 359             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 360             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 361
 362             }
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             if (gmx_mm_any_lt(rsq30,rcutoff2))
 369             {
 370
 371             r30              = _mm_mul_pd(rsq30,rinv30);
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq30             = _mm_mul_pd(iq3,jq0);
 375
 376             /* EWALD ELECTROSTATICS */
 377
 378             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 379             ewrt             = _mm_mul_pd(r30,ewtabscale);
 380             ewitab           = _mm_cvttpd_epi32(ewrt);
 381 #ifdef __XOP__
 382             eweps            = _mm_frcz_pd(ewrt);
 383 #else
 384             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 385 #endif
 386             twoeweps         = _mm_add_pd(eweps,eweps);
 387             ewitab           = _mm_slli_epi32(ewitab,2);
 388             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 389             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 390             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 391             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 392             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 393             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 394             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 395             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 396             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 397             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 398
 399             d                = _mm_sub_pd(r30,rswitch);
 400             d                = _mm_max_pd(d,_mm_setzero_pd());
 401             d2               = _mm_mul_pd(d,d);
 402             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 403
 404             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 405
 406             /* Evaluate switch function */
 407             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 408             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 409             velec            = _mm_mul_pd(velec,sw);
 410             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 411
 412             /* Update potential sum for this i atom from the interaction with this j atom. */
 413             velec            = _mm_and_pd(velec,cutoff_mask);
 414             velecsum         = _mm_add_pd(velecsum,velec);
 415
 416             fscal            = felec;
 417
 418             fscal            = _mm_and_pd(fscal,cutoff_mask);
 419
 420             /* Update vectorial force */
 421             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 422             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 423             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 424
 425             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 426             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 427             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 428
 429             }
 430
 431             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 432
 433             /* Inner loop uses 207 flops */
 434         }
 435
 436         if(jidx<j_index_end)
 437         {
 438
 439             jnrA             = jjnr[jidx];
 440             j_coord_offsetA  = DIM*jnrA;
 441
 442             /* load j atom coordinates */
 443             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx10             = _mm_sub_pd(ix1,jx0);
 448             dy10             = _mm_sub_pd(iy1,jy0);
 449             dz10             = _mm_sub_pd(iz1,jz0);
 450             dx20             = _mm_sub_pd(ix2,jx0);
 451             dy20             = _mm_sub_pd(iy2,jy0);
 452             dz20             = _mm_sub_pd(iz2,jz0);
 453             dx30             = _mm_sub_pd(ix3,jx0);
 454             dy30             = _mm_sub_pd(iy3,jy0);
 455             dz30             = _mm_sub_pd(iz3,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 459             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 460             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 461
 462             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 463             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 464             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 465
 466             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 467             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 468             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 469
 470             /* Load parameters for j particles */
 471             jq0              = _mm_load_sd(charge+jnrA+0);
 472
 473             fjx0             = _mm_setzero_pd();
 474             fjy0             = _mm_setzero_pd();
 475             fjz0             = _mm_setzero_pd();
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             if (gmx_mm_any_lt(rsq10,rcutoff2))
 482             {
 483
 484             r10              = _mm_mul_pd(rsq10,rinv10);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq10             = _mm_mul_pd(iq1,jq0);
 488
 489             /* EWALD ELECTROSTATICS */
 490
 491             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 492             ewrt             = _mm_mul_pd(r10,ewtabscale);
 493             ewitab           = _mm_cvttpd_epi32(ewrt);
 494 #ifdef __XOP__
 495             eweps            = _mm_frcz_pd(ewrt);
 496 #else
 497             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 498 #endif
 499             twoeweps         = _mm_add_pd(eweps,eweps);
 500             ewitab           = _mm_slli_epi32(ewitab,2);
 501             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 502             ewtabD           = _mm_setzero_pd();
 503             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 504             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 505             ewtabFn          = _mm_setzero_pd();
 506             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 507             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 508             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 509             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 510             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 511
 512             d                = _mm_sub_pd(r10,rswitch);
 513             d                = _mm_max_pd(d,_mm_setzero_pd());
 514             d2               = _mm_mul_pd(d,d);
 515             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 516
 517             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 518
 519             /* Evaluate switch function */
 520             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 521             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 522             velec            = _mm_mul_pd(velec,sw);
 523             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             velec            = _mm_and_pd(velec,cutoff_mask);
 527             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 528             velecsum         = _mm_add_pd(velecsum,velec);
 529
 530             fscal            = felec;
 531
 532             fscal            = _mm_and_pd(fscal,cutoff_mask);
 533
 534             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 535
 536             /* Update vectorial force */
 537             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 538             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 539             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 540
 541             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 542             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 543             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 544
 545             }
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             if (gmx_mm_any_lt(rsq20,rcutoff2))
 552             {
 553
 554             r20              = _mm_mul_pd(rsq20,rinv20);
 555
 556             /* Compute parameters for interactions between i and j atoms */
 557             qq20             = _mm_mul_pd(iq2,jq0);
 558
 559             /* EWALD ELECTROSTATICS */
 560
 561             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 562             ewrt             = _mm_mul_pd(r20,ewtabscale);
 563             ewitab           = _mm_cvttpd_epi32(ewrt);
 564 #ifdef __XOP__
 565             eweps            = _mm_frcz_pd(ewrt);
 566 #else
 567             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 568 #endif
 569             twoeweps         = _mm_add_pd(eweps,eweps);
 570             ewitab           = _mm_slli_epi32(ewitab,2);
 571             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 572             ewtabD           = _mm_setzero_pd();
 573             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 574             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 575             ewtabFn          = _mm_setzero_pd();
 576             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 577             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 578             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 579             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 580             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 581
 582             d                = _mm_sub_pd(r20,rswitch);
 583             d                = _mm_max_pd(d,_mm_setzero_pd());
 584             d2               = _mm_mul_pd(d,d);
 585             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 586
 587             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 588
 589             /* Evaluate switch function */
 590             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 591             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 592             velec            = _mm_mul_pd(velec,sw);
 593             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 594
 595             /* Update potential sum for this i atom from the interaction with this j atom. */
 596             velec            = _mm_and_pd(velec,cutoff_mask);
 597             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 598             velecsum         = _mm_add_pd(velecsum,velec);
 599
 600             fscal            = felec;
 601
 602             fscal            = _mm_and_pd(fscal,cutoff_mask);
 603
 604             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 605
 606             /* Update vectorial force */
 607             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 608             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 609             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 610
 611             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 612             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 613             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 614
 615             }
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             if (gmx_mm_any_lt(rsq30,rcutoff2))
 622             {
 623
 624             r30              = _mm_mul_pd(rsq30,rinv30);
 625
 626             /* Compute parameters for interactions between i and j atoms */
 627             qq30             = _mm_mul_pd(iq3,jq0);
 628
 629             /* EWALD ELECTROSTATICS */
 630
 631             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 632             ewrt             = _mm_mul_pd(r30,ewtabscale);
 633             ewitab           = _mm_cvttpd_epi32(ewrt);
 634 #ifdef __XOP__
 635             eweps            = _mm_frcz_pd(ewrt);
 636 #else
 637             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 638 #endif
 639             twoeweps         = _mm_add_pd(eweps,eweps);
 640             ewitab           = _mm_slli_epi32(ewitab,2);
 641             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 642             ewtabD           = _mm_setzero_pd();
 643             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 644             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 645             ewtabFn          = _mm_setzero_pd();
 646             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 647             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 648             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 649             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 650             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 651
 652             d                = _mm_sub_pd(r30,rswitch);
 653             d                = _mm_max_pd(d,_mm_setzero_pd());
 654             d2               = _mm_mul_pd(d,d);
 655             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 656
 657             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 658
 659             /* Evaluate switch function */
 660             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 661             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 662             velec            = _mm_mul_pd(velec,sw);
 663             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 664
 665             /* Update potential sum for this i atom from the interaction with this j atom. */
 666             velec            = _mm_and_pd(velec,cutoff_mask);
 667             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 668             velecsum         = _mm_add_pd(velecsum,velec);
 669
 670             fscal            = felec;
 671
 672             fscal            = _mm_and_pd(fscal,cutoff_mask);
 673
 674             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 675
 676             /* Update vectorial force */
 677             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 678             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 679             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 680
 681             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 682             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 683             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 684
 685             }
 686
 687             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 688
 689             /* Inner loop uses 207 flops */
 690         }
 691
 692         /* End of innermost loop */
 693
 694         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 695                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 696
 697         ggid                        = gid[iidx];
 698         /* Update potential energies */
 699         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 700
 701         /* Increment number of inner iterations */
 702         inneriter                  += j_index_end - j_index_start;
 703
 704         /* Outer loop uses 19 flops */
 705     }
 706
 707     /* Increment number of outer iterations */
 708     outeriter        += nri;
 709
 710     /* Update outer/inner flops */
 711
 712     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*207);
 713 }
 714 /*
 715  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_avx_128_fma_double
 716  * Electrostatics interaction: Ewald
 717  * VdW interaction:            None
 718  * Geometry:                   Water4-Particle
 719  * Calculate force/pot:        Force
 720  */
 721 void
 722 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_avx_128_fma_double
 723                     (t_nblist                    * gmx_restrict       nlist,
 724                      rvec                        * gmx_restrict          xx,
 725                      rvec                        * gmx_restrict          ff,
 726                      t_forcerec                  * gmx_restrict          fr,
 727                      t_mdatoms                   * gmx_restrict     mdatoms,
 728                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 729                      t_nrnb                      * gmx_restrict        nrnb)
 730 {
 731     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 732      * just 0 for non-waters.
 733      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 734      * jnr indices corresponding to data put in the four positions in the SIMD register.
 735      */
 736     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 737     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 738     int              jnrA,jnrB;
 739     int              j_coord_offsetA,j_coord_offsetB;
 740     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 741     real             rcutoff_scalar;
 742     real             *shiftvec,*fshift,*x,*f;
 743     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 744     int              vdwioffset1;
 745     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 746     int              vdwioffset2;
 747     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 748     int              vdwioffset3;
 749     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 750     int              vdwjidx0A,vdwjidx0B;
 751     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 752     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 753     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 754     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 755     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 756     real             *charge;
 757     __m128i          ewitab;
 758     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 759     real             *ewtab;
 760     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 761     real             rswitch_scalar,d_scalar;
 762     __m128d          dummy_mask,cutoff_mask;
 763     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 764     __m128d          one     = _mm_set1_pd(1.0);
 765     __m128d          two     = _mm_set1_pd(2.0);
 766     x                = xx[0];
 767     f                = ff[0];
 768
 769     nri              = nlist->nri;
 770     iinr             = nlist->iinr;
 771     jindex           = nlist->jindex;
 772     jjnr             = nlist->jjnr;
 773     shiftidx         = nlist->shift;
 774     gid              = nlist->gid;
 775     shiftvec         = fr->shift_vec[0];
 776     fshift           = fr->fshift[0];
 777     facel            = _mm_set1_pd(fr->epsfac);
 778     charge           = mdatoms->chargeA;
 779
 780     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 781     ewtab            = fr->ic->tabq_coul_FDV0;
 782     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 783     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 784
 785     /* Setup water-specific parameters */
 786     inr              = nlist->iinr[0];
 787     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 788     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 789     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 790
 791     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 792     rcutoff_scalar   = fr->rcoulomb;
 793     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 794     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 795
 796     rswitch_scalar   = fr->rcoulomb_switch;
 797     rswitch          = _mm_set1_pd(rswitch_scalar);
 798     /* Setup switch parameters */
 799     d_scalar         = rcutoff_scalar-rswitch_scalar;
 800     d                = _mm_set1_pd(d_scalar);
 801     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 802     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 803     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 804     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 805     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 806     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 807
 808     /* Avoid stupid compiler warnings */
 809     jnrA = jnrB = 0;
 810     j_coord_offsetA = 0;
 811     j_coord_offsetB = 0;
 812
 813     outeriter        = 0;
 814     inneriter        = 0;
 815
 816     /* Start outer loop over neighborlists */
 817     for(iidx=0; iidx<nri; iidx++)
 818     {
 819         /* Load shift vector for this list */
 820         i_shift_offset   = DIM*shiftidx[iidx];
 821
 822         /* Load limits for loop over neighbors */
 823         j_index_start    = jindex[iidx];
 824         j_index_end      = jindex[iidx+1];
 825
 826         /* Get outer coordinate index */
 827         inr              = iinr[iidx];
 828         i_coord_offset   = DIM*inr;
 829
 830         /* Load i particle coords and add shift vector */
 831         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 832                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 833
 834         fix1             = _mm_setzero_pd();
 835         fiy1             = _mm_setzero_pd();
 836         fiz1             = _mm_setzero_pd();
 837         fix2             = _mm_setzero_pd();
 838         fiy2             = _mm_setzero_pd();
 839         fiz2             = _mm_setzero_pd();
 840         fix3             = _mm_setzero_pd();
 841         fiy3             = _mm_setzero_pd();
 842         fiz3             = _mm_setzero_pd();
 843
 844         /* Start inner kernel loop */
 845         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 846         {
 847
 848             /* Get j neighbor index, and coordinate index */
 849             jnrA             = jjnr[jidx];
 850             jnrB             = jjnr[jidx+1];
 851             j_coord_offsetA  = DIM*jnrA;
 852             j_coord_offsetB  = DIM*jnrB;
 853
 854             /* load j atom coordinates */
 855             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 856                                               &jx0,&jy0,&jz0);
 857
 858             /* Calculate displacement vector */
 859             dx10             = _mm_sub_pd(ix1,jx0);
 860             dy10             = _mm_sub_pd(iy1,jy0);
 861             dz10             = _mm_sub_pd(iz1,jz0);
 862             dx20             = _mm_sub_pd(ix2,jx0);
 863             dy20             = _mm_sub_pd(iy2,jy0);
 864             dz20             = _mm_sub_pd(iz2,jz0);
 865             dx30             = _mm_sub_pd(ix3,jx0);
 866             dy30             = _mm_sub_pd(iy3,jy0);
 867             dz30             = _mm_sub_pd(iz3,jz0);
 868
 869             /* Calculate squared distance and things based on it */
 870             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 871             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 872             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 873
 874             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 875             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 876             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 877
 878             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 879             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 880             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 881
 882             /* Load parameters for j particles */
 883             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 884
 885             fjx0             = _mm_setzero_pd();
 886             fjy0             = _mm_setzero_pd();
 887             fjz0             = _mm_setzero_pd();
 888
 889             /**************************
 890              * CALCULATE INTERACTIONS *
 891              **************************/
 892
 893             if (gmx_mm_any_lt(rsq10,rcutoff2))
 894             {
 895
 896             r10              = _mm_mul_pd(rsq10,rinv10);
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             qq10             = _mm_mul_pd(iq1,jq0);
 900
 901             /* EWALD ELECTROSTATICS */
 902
 903             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 904             ewrt             = _mm_mul_pd(r10,ewtabscale);
 905             ewitab           = _mm_cvttpd_epi32(ewrt);
 906 #ifdef __XOP__
 907             eweps            = _mm_frcz_pd(ewrt);
 908 #else
 909             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 910 #endif
 911             twoeweps         = _mm_add_pd(eweps,eweps);
 912             ewitab           = _mm_slli_epi32(ewitab,2);
 913             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 914             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 915             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 916             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 917             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 918             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 919             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 920             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 921             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 922             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 923
 924             d                = _mm_sub_pd(r10,rswitch);
 925             d                = _mm_max_pd(d,_mm_setzero_pd());
 926             d2               = _mm_mul_pd(d,d);
 927             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 928
 929             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 930
 931             /* Evaluate switch function */
 932             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 933             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 934             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 935
 936             fscal            = felec;
 937
 938             fscal            = _mm_and_pd(fscal,cutoff_mask);
 939
 940             /* Update vectorial force */
 941             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 942             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 943             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 944
 945             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 946             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 947             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 948
 949             }
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             if (gmx_mm_any_lt(rsq20,rcutoff2))
 956             {
 957
 958             r20              = _mm_mul_pd(rsq20,rinv20);
 959
 960             /* Compute parameters for interactions between i and j atoms */
 961             qq20             = _mm_mul_pd(iq2,jq0);
 962
 963             /* EWALD ELECTROSTATICS */
 964
 965             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 966             ewrt             = _mm_mul_pd(r20,ewtabscale);
 967             ewitab           = _mm_cvttpd_epi32(ewrt);
 968 #ifdef __XOP__
 969             eweps            = _mm_frcz_pd(ewrt);
 970 #else
 971             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 972 #endif
 973             twoeweps         = _mm_add_pd(eweps,eweps);
 974             ewitab           = _mm_slli_epi32(ewitab,2);
 975             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 976             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 977             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 978             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 979             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 980             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 981             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 982             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 983             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 984             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 985
 986             d                = _mm_sub_pd(r20,rswitch);
 987             d                = _mm_max_pd(d,_mm_setzero_pd());
 988             d2               = _mm_mul_pd(d,d);
 989             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 990
 991             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 992
 993             /* Evaluate switch function */
 994             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 995             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 996             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 997
 998             fscal            = felec;
 999
1000             fscal            = _mm_and_pd(fscal,cutoff_mask);
1001
1002             /* Update vectorial force */
1003             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1004             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1005             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1006
1007             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1008             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1009             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1010
1011             }
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             if (gmx_mm_any_lt(rsq30,rcutoff2))
1018             {
1019
1020             r30              = _mm_mul_pd(rsq30,rinv30);
1021
1022             /* Compute parameters for interactions between i and j atoms */
1023             qq30             = _mm_mul_pd(iq3,jq0);
1024
1025             /* EWALD ELECTROSTATICS */
1026
1027             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1028             ewrt             = _mm_mul_pd(r30,ewtabscale);
1029             ewitab           = _mm_cvttpd_epi32(ewrt);
1030 #ifdef __XOP__
1031             eweps            = _mm_frcz_pd(ewrt);
1032 #else
1033             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1034 #endif
1035             twoeweps         = _mm_add_pd(eweps,eweps);
1036             ewitab           = _mm_slli_epi32(ewitab,2);
1037             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1038             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1039             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1040             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1041             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1042             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1043             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1044             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1045             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1046             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1047
1048             d                = _mm_sub_pd(r30,rswitch);
1049             d                = _mm_max_pd(d,_mm_setzero_pd());
1050             d2               = _mm_mul_pd(d,d);
1051             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1052
1053             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1054
1055             /* Evaluate switch function */
1056             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1057             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1058             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1059
1060             fscal            = felec;
1061
1062             fscal            = _mm_and_pd(fscal,cutoff_mask);
1063
1064             /* Update vectorial force */
1065             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1066             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1067             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1068
1069             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1070             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1071             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1072
1073             }
1074
1075             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1076
1077             /* Inner loop uses 198 flops */
1078         }
1079
1080         if(jidx<j_index_end)
1081         {
1082
1083             jnrA             = jjnr[jidx];
1084             j_coord_offsetA  = DIM*jnrA;
1085
1086             /* load j atom coordinates */
1087             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1088                                               &jx0,&jy0,&jz0);
1089
1090             /* Calculate displacement vector */
1091             dx10             = _mm_sub_pd(ix1,jx0);
1092             dy10             = _mm_sub_pd(iy1,jy0);
1093             dz10             = _mm_sub_pd(iz1,jz0);
1094             dx20             = _mm_sub_pd(ix2,jx0);
1095             dy20             = _mm_sub_pd(iy2,jy0);
1096             dz20             = _mm_sub_pd(iz2,jz0);
1097             dx30             = _mm_sub_pd(ix3,jx0);
1098             dy30             = _mm_sub_pd(iy3,jy0);
1099             dz30             = _mm_sub_pd(iz3,jz0);
1100
1101             /* Calculate squared distance and things based on it */
1102             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1103             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1104             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1105
1106             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1107             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1108             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1109
1110             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1111             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1112             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1113
1114             /* Load parameters for j particles */
1115             jq0              = _mm_load_sd(charge+jnrA+0);
1116
1117             fjx0             = _mm_setzero_pd();
1118             fjy0             = _mm_setzero_pd();
1119             fjz0             = _mm_setzero_pd();
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             if (gmx_mm_any_lt(rsq10,rcutoff2))
1126             {
1127
1128             r10              = _mm_mul_pd(rsq10,rinv10);
1129
1130             /* Compute parameters for interactions between i and j atoms */
1131             qq10             = _mm_mul_pd(iq1,jq0);
1132
1133             /* EWALD ELECTROSTATICS */
1134
1135             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1136             ewrt             = _mm_mul_pd(r10,ewtabscale);
1137             ewitab           = _mm_cvttpd_epi32(ewrt);
1138 #ifdef __XOP__
1139             eweps            = _mm_frcz_pd(ewrt);
1140 #else
1141             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1142 #endif
1143             twoeweps         = _mm_add_pd(eweps,eweps);
1144             ewitab           = _mm_slli_epi32(ewitab,2);
1145             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1146             ewtabD           = _mm_setzero_pd();
1147             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1148             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1149             ewtabFn          = _mm_setzero_pd();
1150             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1151             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1152             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1153             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1154             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1155
1156             d                = _mm_sub_pd(r10,rswitch);
1157             d                = _mm_max_pd(d,_mm_setzero_pd());
1158             d2               = _mm_mul_pd(d,d);
1159             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1160
1161             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1162
1163             /* Evaluate switch function */
1164             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1165             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1166             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1167
1168             fscal            = felec;
1169
1170             fscal            = _mm_and_pd(fscal,cutoff_mask);
1171
1172             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1173
1174             /* Update vectorial force */
1175             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1176             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1177             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1178
1179             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1180             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1181             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1182
1183             }
1184
1185             /**************************
1186              * CALCULATE INTERACTIONS *
1187              **************************/
1188
1189             if (gmx_mm_any_lt(rsq20,rcutoff2))
1190             {
1191
1192             r20              = _mm_mul_pd(rsq20,rinv20);
1193
1194             /* Compute parameters for interactions between i and j atoms */
1195             qq20             = _mm_mul_pd(iq2,jq0);
1196
1197             /* EWALD ELECTROSTATICS */
1198
1199             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1200             ewrt             = _mm_mul_pd(r20,ewtabscale);
1201             ewitab           = _mm_cvttpd_epi32(ewrt);
1202 #ifdef __XOP__
1203             eweps            = _mm_frcz_pd(ewrt);
1204 #else
1205             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1206 #endif
1207             twoeweps         = _mm_add_pd(eweps,eweps);
1208             ewitab           = _mm_slli_epi32(ewitab,2);
1209             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1210             ewtabD           = _mm_setzero_pd();
1211             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1212             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1213             ewtabFn          = _mm_setzero_pd();
1214             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1215             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1216             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1217             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1218             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1219
1220             d                = _mm_sub_pd(r20,rswitch);
1221             d                = _mm_max_pd(d,_mm_setzero_pd());
1222             d2               = _mm_mul_pd(d,d);
1223             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1224
1225             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1226
1227             /* Evaluate switch function */
1228             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1229             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1230             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1231
1232             fscal            = felec;
1233
1234             fscal            = _mm_and_pd(fscal,cutoff_mask);
1235
1236             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1237
1238             /* Update vectorial force */
1239             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1240             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1241             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1242
1243             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1244             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1245             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1246
1247             }
1248
1249             /**************************
1250              * CALCULATE INTERACTIONS *
1251              **************************/
1252
1253             if (gmx_mm_any_lt(rsq30,rcutoff2))
1254             {
1255
1256             r30              = _mm_mul_pd(rsq30,rinv30);
1257
1258             /* Compute parameters for interactions between i and j atoms */
1259             qq30             = _mm_mul_pd(iq3,jq0);
1260
1261             /* EWALD ELECTROSTATICS */
1262
1263             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1264             ewrt             = _mm_mul_pd(r30,ewtabscale);
1265             ewitab           = _mm_cvttpd_epi32(ewrt);
1266 #ifdef __XOP__
1267             eweps            = _mm_frcz_pd(ewrt);
1268 #else
1269             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1270 #endif
1271             twoeweps         = _mm_add_pd(eweps,eweps);
1272             ewitab           = _mm_slli_epi32(ewitab,2);
1273             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1274             ewtabD           = _mm_setzero_pd();
1275             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1276             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1277             ewtabFn          = _mm_setzero_pd();
1278             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1279             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1280             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1281             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1282             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1283
1284             d                = _mm_sub_pd(r30,rswitch);
1285             d                = _mm_max_pd(d,_mm_setzero_pd());
1286             d2               = _mm_mul_pd(d,d);
1287             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1288
1289             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1290
1291             /* Evaluate switch function */
1292             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1293             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1294             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1295
1296             fscal            = felec;
1297
1298             fscal            = _mm_and_pd(fscal,cutoff_mask);
1299
1300             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1301
1302             /* Update vectorial force */
1303             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1304             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1305             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1306
1307             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1308             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1309             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1310
1311             }
1312
1313             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1314
1315             /* Inner loop uses 198 flops */
1316         }
1317
1318         /* End of innermost loop */
1319
1320         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1321                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1322
1323         /* Increment number of inner iterations */
1324         inneriter                  += j_index_end - j_index_start;
1325
1326         /* Outer loop uses 18 flops */
1327     }
1328
1329     /* Increment number of outer iterations */
1330     outeriter        += nri;
1331
1332     /* Update outer/inner flops */
1333
1334     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*198);
1335 }