src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSw_VdwNone_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  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              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m128i          ewitab;
  94     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     real             *ewtab;
  96     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  97     real             rswitch_scalar,d_scalar;
  98     __m128d          dummy_mask,cutoff_mask;
  99     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 100     __m128d          one     = _mm_set1_pd(1.0);
 101     __m128d          two     = _mm_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115
 116     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 117     ewtab            = fr->ic->tabq_coul_FDV0;
 118     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 119     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 124     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 125     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 126
 127     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 128     rcutoff_scalar   = fr->rcoulomb;
 129     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 130     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 131
 132     rswitch_scalar   = fr->rcoulomb_switch;
 133     rswitch          = _mm_set1_pd(rswitch_scalar);
 134     /* Setup switch parameters */
 135     d_scalar         = rcutoff_scalar-rswitch_scalar;
 136     d                = _mm_set1_pd(d_scalar);
 137     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 138     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 139     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 140     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 141     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 142     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_pd();
 171         fiy0             = _mm_setzero_pd();
 172         fiz0             = _mm_setzero_pd();
 173         fix1             = _mm_setzero_pd();
 174         fiy1             = _mm_setzero_pd();
 175         fiz1             = _mm_setzero_pd();
 176         fix2             = _mm_setzero_pd();
 177         fiy2             = _mm_setzero_pd();
 178         fiz2             = _mm_setzero_pd();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_pd();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_pd(ix0,jx0);
 199             dy00             = _mm_sub_pd(iy0,jy0);
 200             dz00             = _mm_sub_pd(iz0,jz0);
 201             dx10             = _mm_sub_pd(ix1,jx0);
 202             dy10             = _mm_sub_pd(iy1,jy0);
 203             dz10             = _mm_sub_pd(iz1,jz0);
 204             dx20             = _mm_sub_pd(ix2,jx0);
 205             dy20             = _mm_sub_pd(iy2,jy0);
 206             dz20             = _mm_sub_pd(iz2,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 210             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 211             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 212
 213             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 214             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 215             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 216
 217             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 218             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 219             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 223
 224             fjx0             = _mm_setzero_pd();
 225             fjy0             = _mm_setzero_pd();
 226             fjz0             = _mm_setzero_pd();
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             if (gmx_mm_any_lt(rsq00,rcutoff2))
 233             {
 234
 235             r00              = _mm_mul_pd(rsq00,rinv00);
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             qq00             = _mm_mul_pd(iq0,jq0);
 239
 240             /* EWALD ELECTROSTATICS */
 241
 242             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 243             ewrt             = _mm_mul_pd(r00,ewtabscale);
 244             ewitab           = _mm_cvttpd_epi32(ewrt);
 245 #ifdef __XOP__
 246             eweps            = _mm_frcz_pd(ewrt);
 247 #else
 248             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 249 #endif
 250             twoeweps         = _mm_add_pd(eweps,eweps);
 251             ewitab           = _mm_slli_epi32(ewitab,2);
 252             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 253             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 254             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 255             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 256             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 257             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 258             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 259             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 260             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 261             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 262
 263             d                = _mm_sub_pd(r00,rswitch);
 264             d                = _mm_max_pd(d,_mm_setzero_pd());
 265             d2               = _mm_mul_pd(d,d);
 266             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 267
 268             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 269
 270             /* Evaluate switch function */
 271             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 272             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 273             velec            = _mm_mul_pd(velec,sw);
 274             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velec            = _mm_and_pd(velec,cutoff_mask);
 278             velecsum         = _mm_add_pd(velecsum,velec);
 279
 280             fscal            = felec;
 281
 282             fscal            = _mm_and_pd(fscal,cutoff_mask);
 283
 284             /* Update vectorial force */
 285             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 286             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 287             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 288
 289             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 290             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 291             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 292
 293             }
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             if (gmx_mm_any_lt(rsq10,rcutoff2))
 300             {
 301
 302             r10              = _mm_mul_pd(rsq10,rinv10);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm_mul_pd(iq1,jq0);
 306
 307             /* EWALD ELECTROSTATICS */
 308
 309             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 310             ewrt             = _mm_mul_pd(r10,ewtabscale);
 311             ewitab           = _mm_cvttpd_epi32(ewrt);
 312 #ifdef __XOP__
 313             eweps            = _mm_frcz_pd(ewrt);
 314 #else
 315             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 316 #endif
 317             twoeweps         = _mm_add_pd(eweps,eweps);
 318             ewitab           = _mm_slli_epi32(ewitab,2);
 319             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 320             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 321             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 322             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 323             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 324             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 325             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 326             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 327             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 328             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 329
 330             d                = _mm_sub_pd(r10,rswitch);
 331             d                = _mm_max_pd(d,_mm_setzero_pd());
 332             d2               = _mm_mul_pd(d,d);
 333             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 334
 335             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 336
 337             /* Evaluate switch function */
 338             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 339             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 340             velec            = _mm_mul_pd(velec,sw);
 341             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velec            = _mm_and_pd(velec,cutoff_mask);
 345             velecsum         = _mm_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             fscal            = _mm_and_pd(fscal,cutoff_mask);
 350
 351             /* Update vectorial force */
 352             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 353             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 354             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 355
 356             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 357             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 358             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 359
 360             }
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             if (gmx_mm_any_lt(rsq20,rcutoff2))
 367             {
 368
 369             r20              = _mm_mul_pd(rsq20,rinv20);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq20             = _mm_mul_pd(iq2,jq0);
 373
 374             /* EWALD ELECTROSTATICS */
 375
 376             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 377             ewrt             = _mm_mul_pd(r20,ewtabscale);
 378             ewitab           = _mm_cvttpd_epi32(ewrt);
 379 #ifdef __XOP__
 380             eweps            = _mm_frcz_pd(ewrt);
 381 #else
 382             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 383 #endif
 384             twoeweps         = _mm_add_pd(eweps,eweps);
 385             ewitab           = _mm_slli_epi32(ewitab,2);
 386             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 387             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 388             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 389             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 390             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 391             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 392             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 393             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 394             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 395             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 396
 397             d                = _mm_sub_pd(r20,rswitch);
 398             d                = _mm_max_pd(d,_mm_setzero_pd());
 399             d2               = _mm_mul_pd(d,d);
 400             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 401
 402             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 403
 404             /* Evaluate switch function */
 405             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 406             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 407             velec            = _mm_mul_pd(velec,sw);
 408             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 409
 410             /* Update potential sum for this i atom from the interaction with this j atom. */
 411             velec            = _mm_and_pd(velec,cutoff_mask);
 412             velecsum         = _mm_add_pd(velecsum,velec);
 413
 414             fscal            = felec;
 415
 416             fscal            = _mm_and_pd(fscal,cutoff_mask);
 417
 418             /* Update vectorial force */
 419             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 420             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 421             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 422
 423             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 424             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 425             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 426
 427             }
 428
 429             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 430
 431             /* Inner loop uses 207 flops */
 432         }
 433
 434         if(jidx<j_index_end)
 435         {
 436
 437             jnrA             = jjnr[jidx];
 438             j_coord_offsetA  = DIM*jnrA;
 439
 440             /* load j atom coordinates */
 441             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 442                                               &jx0,&jy0,&jz0);
 443
 444             /* Calculate displacement vector */
 445             dx00             = _mm_sub_pd(ix0,jx0);
 446             dy00             = _mm_sub_pd(iy0,jy0);
 447             dz00             = _mm_sub_pd(iz0,jz0);
 448             dx10             = _mm_sub_pd(ix1,jx0);
 449             dy10             = _mm_sub_pd(iy1,jy0);
 450             dz10             = _mm_sub_pd(iz1,jz0);
 451             dx20             = _mm_sub_pd(ix2,jx0);
 452             dy20             = _mm_sub_pd(iy2,jy0);
 453             dz20             = _mm_sub_pd(iz2,jz0);
 454
 455             /* Calculate squared distance and things based on it */
 456             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 457             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 458             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 459
 460             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 461             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 462             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 463
 464             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 465             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 466             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 467
 468             /* Load parameters for j particles */
 469             jq0              = _mm_load_sd(charge+jnrA+0);
 470
 471             fjx0             = _mm_setzero_pd();
 472             fjy0             = _mm_setzero_pd();
 473             fjz0             = _mm_setzero_pd();
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             if (gmx_mm_any_lt(rsq00,rcutoff2))
 480             {
 481
 482             r00              = _mm_mul_pd(rsq00,rinv00);
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             qq00             = _mm_mul_pd(iq0,jq0);
 486
 487             /* EWALD ELECTROSTATICS */
 488
 489             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 490             ewrt             = _mm_mul_pd(r00,ewtabscale);
 491             ewitab           = _mm_cvttpd_epi32(ewrt);
 492 #ifdef __XOP__
 493             eweps            = _mm_frcz_pd(ewrt);
 494 #else
 495             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 496 #endif
 497             twoeweps         = _mm_add_pd(eweps,eweps);
 498             ewitab           = _mm_slli_epi32(ewitab,2);
 499             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 500             ewtabD           = _mm_setzero_pd();
 501             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 502             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 503             ewtabFn          = _mm_setzero_pd();
 504             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 505             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 506             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 507             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 508             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 509
 510             d                = _mm_sub_pd(r00,rswitch);
 511             d                = _mm_max_pd(d,_mm_setzero_pd());
 512             d2               = _mm_mul_pd(d,d);
 513             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 514
 515             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 516
 517             /* Evaluate switch function */
 518             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 519             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 520             velec            = _mm_mul_pd(velec,sw);
 521             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             velec            = _mm_and_pd(velec,cutoff_mask);
 525             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 526             velecsum         = _mm_add_pd(velecsum,velec);
 527
 528             fscal            = felec;
 529
 530             fscal            = _mm_and_pd(fscal,cutoff_mask);
 531
 532             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 533
 534             /* Update vectorial force */
 535             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 536             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 537             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 538
 539             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 540             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 541             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 542
 543             }
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             if (gmx_mm_any_lt(rsq10,rcutoff2))
 550             {
 551
 552             r10              = _mm_mul_pd(rsq10,rinv10);
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq10             = _mm_mul_pd(iq1,jq0);
 556
 557             /* EWALD ELECTROSTATICS */
 558
 559             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 560             ewrt             = _mm_mul_pd(r10,ewtabscale);
 561             ewitab           = _mm_cvttpd_epi32(ewrt);
 562 #ifdef __XOP__
 563             eweps            = _mm_frcz_pd(ewrt);
 564 #else
 565             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 566 #endif
 567             twoeweps         = _mm_add_pd(eweps,eweps);
 568             ewitab           = _mm_slli_epi32(ewitab,2);
 569             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 570             ewtabD           = _mm_setzero_pd();
 571             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 572             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 573             ewtabFn          = _mm_setzero_pd();
 574             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 575             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 576             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 577             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 578             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 579
 580             d                = _mm_sub_pd(r10,rswitch);
 581             d                = _mm_max_pd(d,_mm_setzero_pd());
 582             d2               = _mm_mul_pd(d,d);
 583             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 584
 585             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 586
 587             /* Evaluate switch function */
 588             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 589             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 590             velec            = _mm_mul_pd(velec,sw);
 591             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm_and_pd(velec,cutoff_mask);
 595             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 596             velecsum         = _mm_add_pd(velecsum,velec);
 597
 598             fscal            = felec;
 599
 600             fscal            = _mm_and_pd(fscal,cutoff_mask);
 601
 602             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 603
 604             /* Update vectorial force */
 605             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 606             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 607             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 608
 609             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 610             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 611             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 612
 613             }
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             if (gmx_mm_any_lt(rsq20,rcutoff2))
 620             {
 621
 622             r20              = _mm_mul_pd(rsq20,rinv20);
 623
 624             /* Compute parameters for interactions between i and j atoms */
 625             qq20             = _mm_mul_pd(iq2,jq0);
 626
 627             /* EWALD ELECTROSTATICS */
 628
 629             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 630             ewrt             = _mm_mul_pd(r20,ewtabscale);
 631             ewitab           = _mm_cvttpd_epi32(ewrt);
 632 #ifdef __XOP__
 633             eweps            = _mm_frcz_pd(ewrt);
 634 #else
 635             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 636 #endif
 637             twoeweps         = _mm_add_pd(eweps,eweps);
 638             ewitab           = _mm_slli_epi32(ewitab,2);
 639             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 640             ewtabD           = _mm_setzero_pd();
 641             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 642             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 643             ewtabFn          = _mm_setzero_pd();
 644             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 645             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 646             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 647             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 648             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 649
 650             d                = _mm_sub_pd(r20,rswitch);
 651             d                = _mm_max_pd(d,_mm_setzero_pd());
 652             d2               = _mm_mul_pd(d,d);
 653             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 654
 655             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 656
 657             /* Evaluate switch function */
 658             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 659             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 660             velec            = _mm_mul_pd(velec,sw);
 661             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 662
 663             /* Update potential sum for this i atom from the interaction with this j atom. */
 664             velec            = _mm_and_pd(velec,cutoff_mask);
 665             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 666             velecsum         = _mm_add_pd(velecsum,velec);
 667
 668             fscal            = felec;
 669
 670             fscal            = _mm_and_pd(fscal,cutoff_mask);
 671
 672             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 673
 674             /* Update vectorial force */
 675             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 676             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 677             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 678
 679             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 680             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 681             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 682
 683             }
 684
 685             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 686
 687             /* Inner loop uses 207 flops */
 688         }
 689
 690         /* End of innermost loop */
 691
 692         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 693                                               f+i_coord_offset,fshift+i_shift_offset);
 694
 695         ggid                        = gid[iidx];
 696         /* Update potential energies */
 697         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 698
 699         /* Increment number of inner iterations */
 700         inneriter                  += j_index_end - j_index_start;
 701
 702         /* Outer loop uses 19 flops */
 703     }
 704
 705     /* Increment number of outer iterations */
 706     outeriter        += nri;
 707
 708     /* Update outer/inner flops */
 709
 710     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*207);
 711 }
 712 /*
 713  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_128_fma_double
 714  * Electrostatics interaction: Ewald
 715  * VdW interaction:            None
 716  * Geometry:                   Water3-Particle
 717  * Calculate force/pot:        Force
 718  */
 719 void
 720 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_128_fma_double
 721                     (t_nblist                    * gmx_restrict       nlist,
 722                      rvec                        * gmx_restrict          xx,
 723                      rvec                        * gmx_restrict          ff,
 724                      t_forcerec                  * gmx_restrict          fr,
 725                      t_mdatoms                   * gmx_restrict     mdatoms,
 726                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 727                      t_nrnb                      * gmx_restrict        nrnb)
 728 {
 729     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 730      * just 0 for non-waters.
 731      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 732      * jnr indices corresponding to data put in the four positions in the SIMD register.
 733      */
 734     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 735     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 736     int              jnrA,jnrB;
 737     int              j_coord_offsetA,j_coord_offsetB;
 738     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 739     real             rcutoff_scalar;
 740     real             *shiftvec,*fshift,*x,*f;
 741     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 742     int              vdwioffset0;
 743     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 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              vdwjidx0A,vdwjidx0B;
 749     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 750     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 751     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 752     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 753     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 754     real             *charge;
 755     __m128i          ewitab;
 756     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 757     real             *ewtab;
 758     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 759     real             rswitch_scalar,d_scalar;
 760     __m128d          dummy_mask,cutoff_mask;
 761     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 762     __m128d          one     = _mm_set1_pd(1.0);
 763     __m128d          two     = _mm_set1_pd(2.0);
 764     x                = xx[0];
 765     f                = ff[0];
 766
 767     nri              = nlist->nri;
 768     iinr             = nlist->iinr;
 769     jindex           = nlist->jindex;
 770     jjnr             = nlist->jjnr;
 771     shiftidx         = nlist->shift;
 772     gid              = nlist->gid;
 773     shiftvec         = fr->shift_vec[0];
 774     fshift           = fr->fshift[0];
 775     facel            = _mm_set1_pd(fr->epsfac);
 776     charge           = mdatoms->chargeA;
 777
 778     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 779     ewtab            = fr->ic->tabq_coul_FDV0;
 780     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 781     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 782
 783     /* Setup water-specific parameters */
 784     inr              = nlist->iinr[0];
 785     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 786     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 787     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 788
 789     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 790     rcutoff_scalar   = fr->rcoulomb;
 791     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 792     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 793
 794     rswitch_scalar   = fr->rcoulomb_switch;
 795     rswitch          = _mm_set1_pd(rswitch_scalar);
 796     /* Setup switch parameters */
 797     d_scalar         = rcutoff_scalar-rswitch_scalar;
 798     d                = _mm_set1_pd(d_scalar);
 799     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 800     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 801     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 802     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 803     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 804     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 805
 806     /* Avoid stupid compiler warnings */
 807     jnrA = jnrB = 0;
 808     j_coord_offsetA = 0;
 809     j_coord_offsetB = 0;
 810
 811     outeriter        = 0;
 812     inneriter        = 0;
 813
 814     /* Start outer loop over neighborlists */
 815     for(iidx=0; iidx<nri; iidx++)
 816     {
 817         /* Load shift vector for this list */
 818         i_shift_offset   = DIM*shiftidx[iidx];
 819
 820         /* Load limits for loop over neighbors */
 821         j_index_start    = jindex[iidx];
 822         j_index_end      = jindex[iidx+1];
 823
 824         /* Get outer coordinate index */
 825         inr              = iinr[iidx];
 826         i_coord_offset   = DIM*inr;
 827
 828         /* Load i particle coords and add shift vector */
 829         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 830                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 831
 832         fix0             = _mm_setzero_pd();
 833         fiy0             = _mm_setzero_pd();
 834         fiz0             = _mm_setzero_pd();
 835         fix1             = _mm_setzero_pd();
 836         fiy1             = _mm_setzero_pd();
 837         fiz1             = _mm_setzero_pd();
 838         fix2             = _mm_setzero_pd();
 839         fiy2             = _mm_setzero_pd();
 840         fiz2             = _mm_setzero_pd();
 841
 842         /* Start inner kernel loop */
 843         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 844         {
 845
 846             /* Get j neighbor index, and coordinate index */
 847             jnrA             = jjnr[jidx];
 848             jnrB             = jjnr[jidx+1];
 849             j_coord_offsetA  = DIM*jnrA;
 850             j_coord_offsetB  = DIM*jnrB;
 851
 852             /* load j atom coordinates */
 853             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 854                                               &jx0,&jy0,&jz0);
 855
 856             /* Calculate displacement vector */
 857             dx00             = _mm_sub_pd(ix0,jx0);
 858             dy00             = _mm_sub_pd(iy0,jy0);
 859             dz00             = _mm_sub_pd(iz0,jz0);
 860             dx10             = _mm_sub_pd(ix1,jx0);
 861             dy10             = _mm_sub_pd(iy1,jy0);
 862             dz10             = _mm_sub_pd(iz1,jz0);
 863             dx20             = _mm_sub_pd(ix2,jx0);
 864             dy20             = _mm_sub_pd(iy2,jy0);
 865             dz20             = _mm_sub_pd(iz2,jz0);
 866
 867             /* Calculate squared distance and things based on it */
 868             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 869             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 870             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 871
 872             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 873             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 874             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 875
 876             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 877             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 878             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 879
 880             /* Load parameters for j particles */
 881             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 882
 883             fjx0             = _mm_setzero_pd();
 884             fjy0             = _mm_setzero_pd();
 885             fjz0             = _mm_setzero_pd();
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             if (gmx_mm_any_lt(rsq00,rcutoff2))
 892             {
 893
 894             r00              = _mm_mul_pd(rsq00,rinv00);
 895
 896             /* Compute parameters for interactions between i and j atoms */
 897             qq00             = _mm_mul_pd(iq0,jq0);
 898
 899             /* EWALD ELECTROSTATICS */
 900
 901             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 902             ewrt             = _mm_mul_pd(r00,ewtabscale);
 903             ewitab           = _mm_cvttpd_epi32(ewrt);
 904 #ifdef __XOP__
 905             eweps            = _mm_frcz_pd(ewrt);
 906 #else
 907             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 908 #endif
 909             twoeweps         = _mm_add_pd(eweps,eweps);
 910             ewitab           = _mm_slli_epi32(ewitab,2);
 911             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 912             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 913             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 914             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 915             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 916             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 917             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 918             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 919             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 920             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 921
 922             d                = _mm_sub_pd(r00,rswitch);
 923             d                = _mm_max_pd(d,_mm_setzero_pd());
 924             d2               = _mm_mul_pd(d,d);
 925             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 926
 927             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 928
 929             /* Evaluate switch function */
 930             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 931             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 932             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 933
 934             fscal            = felec;
 935
 936             fscal            = _mm_and_pd(fscal,cutoff_mask);
 937
 938             /* Update vectorial force */
 939             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 940             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 941             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 942
 943             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 944             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 945             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 946
 947             }
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             if (gmx_mm_any_lt(rsq10,rcutoff2))
 954             {
 955
 956             r10              = _mm_mul_pd(rsq10,rinv10);
 957
 958             /* Compute parameters for interactions between i and j atoms */
 959             qq10             = _mm_mul_pd(iq1,jq0);
 960
 961             /* EWALD ELECTROSTATICS */
 962
 963             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 964             ewrt             = _mm_mul_pd(r10,ewtabscale);
 965             ewitab           = _mm_cvttpd_epi32(ewrt);
 966 #ifdef __XOP__
 967             eweps            = _mm_frcz_pd(ewrt);
 968 #else
 969             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 970 #endif
 971             twoeweps         = _mm_add_pd(eweps,eweps);
 972             ewitab           = _mm_slli_epi32(ewitab,2);
 973             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 974             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 975             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 976             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 977             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 978             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 979             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 980             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 981             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 982             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 983
 984             d                = _mm_sub_pd(r10,rswitch);
 985             d                = _mm_max_pd(d,_mm_setzero_pd());
 986             d2               = _mm_mul_pd(d,d);
 987             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 988
 989             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 990
 991             /* Evaluate switch function */
 992             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 993             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 994             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 995
 996             fscal            = felec;
 997
 998             fscal            = _mm_and_pd(fscal,cutoff_mask);
 999
1000             /* Update vectorial force */
1001             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1002             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1003             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1004
1005             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1006             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1007             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1008
1009             }
1010
1011             /**************************
1012              * CALCULATE INTERACTIONS *
1013              **************************/
1014
1015             if (gmx_mm_any_lt(rsq20,rcutoff2))
1016             {
1017
1018             r20              = _mm_mul_pd(rsq20,rinv20);
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             qq20             = _mm_mul_pd(iq2,jq0);
1022
1023             /* EWALD ELECTROSTATICS */
1024
1025             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1026             ewrt             = _mm_mul_pd(r20,ewtabscale);
1027             ewitab           = _mm_cvttpd_epi32(ewrt);
1028 #ifdef __XOP__
1029             eweps            = _mm_frcz_pd(ewrt);
1030 #else
1031             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1032 #endif
1033             twoeweps         = _mm_add_pd(eweps,eweps);
1034             ewitab           = _mm_slli_epi32(ewitab,2);
1035             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1036             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1037             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1038             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1039             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1040             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1041             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1042             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1043             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1044             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1045
1046             d                = _mm_sub_pd(r20,rswitch);
1047             d                = _mm_max_pd(d,_mm_setzero_pd());
1048             d2               = _mm_mul_pd(d,d);
1049             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1050
1051             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1052
1053             /* Evaluate switch function */
1054             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1055             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1056             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1057
1058             fscal            = felec;
1059
1060             fscal            = _mm_and_pd(fscal,cutoff_mask);
1061
1062             /* Update vectorial force */
1063             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1064             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1065             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1066
1067             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1068             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1069             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1070
1071             }
1072
1073             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1074
1075             /* Inner loop uses 198 flops */
1076         }
1077
1078         if(jidx<j_index_end)
1079         {
1080
1081             jnrA             = jjnr[jidx];
1082             j_coord_offsetA  = DIM*jnrA;
1083
1084             /* load j atom coordinates */
1085             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1086                                               &jx0,&jy0,&jz0);
1087
1088             /* Calculate displacement vector */
1089             dx00             = _mm_sub_pd(ix0,jx0);
1090             dy00             = _mm_sub_pd(iy0,jy0);
1091             dz00             = _mm_sub_pd(iz0,jz0);
1092             dx10             = _mm_sub_pd(ix1,jx0);
1093             dy10             = _mm_sub_pd(iy1,jy0);
1094             dz10             = _mm_sub_pd(iz1,jz0);
1095             dx20             = _mm_sub_pd(ix2,jx0);
1096             dy20             = _mm_sub_pd(iy2,jy0);
1097             dz20             = _mm_sub_pd(iz2,jz0);
1098
1099             /* Calculate squared distance and things based on it */
1100             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1101             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1102             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1103
1104             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1105             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1106             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1107
1108             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1109             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1110             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1111
1112             /* Load parameters for j particles */
1113             jq0              = _mm_load_sd(charge+jnrA+0);
1114
1115             fjx0             = _mm_setzero_pd();
1116             fjy0             = _mm_setzero_pd();
1117             fjz0             = _mm_setzero_pd();
1118
1119             /**************************
1120              * CALCULATE INTERACTIONS *
1121              **************************/
1122
1123             if (gmx_mm_any_lt(rsq00,rcutoff2))
1124             {
1125
1126             r00              = _mm_mul_pd(rsq00,rinv00);
1127
1128             /* Compute parameters for interactions between i and j atoms */
1129             qq00             = _mm_mul_pd(iq0,jq0);
1130
1131             /* EWALD ELECTROSTATICS */
1132
1133             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1134             ewrt             = _mm_mul_pd(r00,ewtabscale);
1135             ewitab           = _mm_cvttpd_epi32(ewrt);
1136 #ifdef __XOP__
1137             eweps            = _mm_frcz_pd(ewrt);
1138 #else
1139             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1140 #endif
1141             twoeweps         = _mm_add_pd(eweps,eweps);
1142             ewitab           = _mm_slli_epi32(ewitab,2);
1143             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1144             ewtabD           = _mm_setzero_pd();
1145             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1146             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1147             ewtabFn          = _mm_setzero_pd();
1148             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1149             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1150             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1151             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
1152             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1153
1154             d                = _mm_sub_pd(r00,rswitch);
1155             d                = _mm_max_pd(d,_mm_setzero_pd());
1156             d2               = _mm_mul_pd(d,d);
1157             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1158
1159             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1160
1161             /* Evaluate switch function */
1162             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1163             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
1164             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1165
1166             fscal            = felec;
1167
1168             fscal            = _mm_and_pd(fscal,cutoff_mask);
1169
1170             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1171
1172             /* Update vectorial force */
1173             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1174             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1175             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1176
1177             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1178             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1179             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1180
1181             }
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             if (gmx_mm_any_lt(rsq10,rcutoff2))
1188             {
1189
1190             r10              = _mm_mul_pd(rsq10,rinv10);
1191
1192             /* Compute parameters for interactions between i and j atoms */
1193             qq10             = _mm_mul_pd(iq1,jq0);
1194
1195             /* EWALD ELECTROSTATICS */
1196
1197             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1198             ewrt             = _mm_mul_pd(r10,ewtabscale);
1199             ewitab           = _mm_cvttpd_epi32(ewrt);
1200 #ifdef __XOP__
1201             eweps            = _mm_frcz_pd(ewrt);
1202 #else
1203             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1204 #endif
1205             twoeweps         = _mm_add_pd(eweps,eweps);
1206             ewitab           = _mm_slli_epi32(ewitab,2);
1207             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1208             ewtabD           = _mm_setzero_pd();
1209             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1210             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1211             ewtabFn          = _mm_setzero_pd();
1212             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1213             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1214             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1215             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1216             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1217
1218             d                = _mm_sub_pd(r10,rswitch);
1219             d                = _mm_max_pd(d,_mm_setzero_pd());
1220             d2               = _mm_mul_pd(d,d);
1221             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1222
1223             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1224
1225             /* Evaluate switch function */
1226             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1227             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1228             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1229
1230             fscal            = felec;
1231
1232             fscal            = _mm_and_pd(fscal,cutoff_mask);
1233
1234             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1235
1236             /* Update vectorial force */
1237             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1238             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1239             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1240
1241             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1242             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1243             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1244
1245             }
1246
1247             /**************************
1248              * CALCULATE INTERACTIONS *
1249              **************************/
1250
1251             if (gmx_mm_any_lt(rsq20,rcutoff2))
1252             {
1253
1254             r20              = _mm_mul_pd(rsq20,rinv20);
1255
1256             /* Compute parameters for interactions between i and j atoms */
1257             qq20             = _mm_mul_pd(iq2,jq0);
1258
1259             /* EWALD ELECTROSTATICS */
1260
1261             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1262             ewrt             = _mm_mul_pd(r20,ewtabscale);
1263             ewitab           = _mm_cvttpd_epi32(ewrt);
1264 #ifdef __XOP__
1265             eweps            = _mm_frcz_pd(ewrt);
1266 #else
1267             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1268 #endif
1269             twoeweps         = _mm_add_pd(eweps,eweps);
1270             ewitab           = _mm_slli_epi32(ewitab,2);
1271             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1272             ewtabD           = _mm_setzero_pd();
1273             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1274             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1275             ewtabFn          = _mm_setzero_pd();
1276             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1277             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1278             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1279             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1280             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1281
1282             d                = _mm_sub_pd(r20,rswitch);
1283             d                = _mm_max_pd(d,_mm_setzero_pd());
1284             d2               = _mm_mul_pd(d,d);
1285             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1286
1287             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1288
1289             /* Evaluate switch function */
1290             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1291             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1292             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1293
1294             fscal            = felec;
1295
1296             fscal            = _mm_and_pd(fscal,cutoff_mask);
1297
1298             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1299
1300             /* Update vectorial force */
1301             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1302             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1303             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1304
1305             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1306             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1307             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1308
1309             }
1310
1311             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1312
1313             /* Inner loop uses 198 flops */
1314         }
1315
1316         /* End of innermost loop */
1317
1318         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1319                                               f+i_coord_offset,fshift+i_shift_offset);
1320
1321         /* Increment number of inner iterations */
1322         inneriter                  += j_index_end - j_index_start;
1323
1324         /* Outer loop uses 18 flops */
1325     }
1326
1327     /* Increment number of outer iterations */
1328     outeriter        += nri;
1329
1330     /* Update outer/inner flops */
1331
1332     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*198);
1333 }