src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse4_1_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     real             *ewtab;
 107     __m128           dummy_mask,cutoff_mask;
 108     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 109     __m128           one     = _mm_set1_ps(1.0);
 110     __m128           two     = _mm_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_ps(fr->ic->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 137     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->ic->rcoulomb;
 142     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 143     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 144
 145     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 146     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = jnrC = jnrD = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152     j_coord_offsetC = 0;
 153     j_coord_offsetD = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 181         fix0             = _mm_setzero_ps();
 182         fiy0             = _mm_setzero_ps();
 183         fiz0             = _mm_setzero_ps();
 184         fix1             = _mm_setzero_ps();
 185         fiy1             = _mm_setzero_ps();
 186         fiz1             = _mm_setzero_ps();
 187         fix2             = _mm_setzero_ps();
 188         fiy2             = _mm_setzero_ps();
 189         fiz2             = _mm_setzero_ps();
 190         fix3             = _mm_setzero_ps();
 191         fiy3             = _mm_setzero_ps();
 192         fiz3             = _mm_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm_setzero_ps();
 196         vvdwsum          = _mm_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                               x+j_coord_offsetC,x+j_coord_offsetD,
 215                                               &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm_sub_ps(ix0,jx0);
 219             dy00             = _mm_sub_ps(iy0,jy0);
 220             dz00             = _mm_sub_ps(iz0,jz0);
 221             dx10             = _mm_sub_ps(ix1,jx0);
 222             dy10             = _mm_sub_ps(iy1,jy0);
 223             dz10             = _mm_sub_ps(iz1,jz0);
 224             dx20             = _mm_sub_ps(ix2,jx0);
 225             dy20             = _mm_sub_ps(iy2,jy0);
 226             dz20             = _mm_sub_ps(iz2,jz0);
 227             dx30             = _mm_sub_ps(ix3,jx0);
 228             dy30             = _mm_sub_ps(iy3,jy0);
 229             dz30             = _mm_sub_ps(iz3,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 233             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 234             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 235             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 236
 237             rinv10           = sse41_invsqrt_f(rsq10);
 238             rinv20           = sse41_invsqrt_f(rsq20);
 239             rinv30           = sse41_invsqrt_f(rsq30);
 240
 241             rinvsq00         = sse41_inv_f(rsq00);
 242             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 243             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 244             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 245
 246             /* Load parameters for j particles */
 247             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 248                                                               charge+jnrC+0,charge+jnrD+0);
 249             vdwjidx0A        = 2*vdwtype[jnrA+0];
 250             vdwjidx0B        = 2*vdwtype[jnrB+0];
 251             vdwjidx0C        = 2*vdwtype[jnrC+0];
 252             vdwjidx0D        = 2*vdwtype[jnrD+0];
 253
 254             fjx0             = _mm_setzero_ps();
 255             fjy0             = _mm_setzero_ps();
 256             fjz0             = _mm_setzero_ps();
 257
 258             /**************************
 259              * CALCULATE INTERACTIONS *
 260              **************************/
 261
 262             if (gmx_mm_any_lt(rsq00,rcutoff2))
 263             {
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 267                                          vdwparam+vdwioffset0+vdwjidx0B,
 268                                          vdwparam+vdwioffset0+vdwjidx0C,
 269                                          vdwparam+vdwioffset0+vdwjidx0D,
 270                                          &c6_00,&c12_00);
 271
 272             /* LENNARD-JONES DISPERSION/REPULSION */
 273
 274             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 275             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 276             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 277             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 278                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 279             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 280
 281             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 285             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 286
 287             fscal            = fvdw;
 288
 289             fscal            = _mm_and_ps(fscal,cutoff_mask);
 290
 291             /* Calculate temporary vectorial force */
 292             tx               = _mm_mul_ps(fscal,dx00);
 293             ty               = _mm_mul_ps(fscal,dy00);
 294             tz               = _mm_mul_ps(fscal,dz00);
 295
 296             /* Update vectorial force */
 297             fix0             = _mm_add_ps(fix0,tx);
 298             fiy0             = _mm_add_ps(fiy0,ty);
 299             fiz0             = _mm_add_ps(fiz0,tz);
 300
 301             fjx0             = _mm_add_ps(fjx0,tx);
 302             fjy0             = _mm_add_ps(fjy0,ty);
 303             fjz0             = _mm_add_ps(fjz0,tz);
 304
 305             }
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             if (gmx_mm_any_lt(rsq10,rcutoff2))
 312             {
 313
 314             r10              = _mm_mul_ps(rsq10,rinv10);
 315
 316             /* Compute parameters for interactions between i and j atoms */
 317             qq10             = _mm_mul_ps(iq1,jq0);
 318
 319             /* EWALD ELECTROSTATICS */
 320
 321             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 322             ewrt             = _mm_mul_ps(r10,ewtabscale);
 323             ewitab           = _mm_cvttps_epi32(ewrt);
 324             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 325             ewitab           = _mm_slli_epi32(ewitab,2);
 326             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 327             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 328             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 329             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 330             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 331             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 332             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 333             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 334             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 335
 336             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm_and_ps(velec,cutoff_mask);
 340             velecsum         = _mm_add_ps(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             fscal            = _mm_and_ps(fscal,cutoff_mask);
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_ps(fscal,dx10);
 348             ty               = _mm_mul_ps(fscal,dy10);
 349             tz               = _mm_mul_ps(fscal,dz10);
 350
 351             /* Update vectorial force */
 352             fix1             = _mm_add_ps(fix1,tx);
 353             fiy1             = _mm_add_ps(fiy1,ty);
 354             fiz1             = _mm_add_ps(fiz1,tz);
 355
 356             fjx0             = _mm_add_ps(fjx0,tx);
 357             fjy0             = _mm_add_ps(fjy0,ty);
 358             fjz0             = _mm_add_ps(fjz0,tz);
 359
 360             }
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             if (gmx_mm_any_lt(rsq20,rcutoff2))
 367             {
 368
 369             r20              = _mm_mul_ps(rsq20,rinv20);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq20             = _mm_mul_ps(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_ps(r20,ewtabscale);
 378             ewitab           = _mm_cvttps_epi32(ewrt);
 379             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 380             ewitab           = _mm_slli_epi32(ewitab,2);
 381             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 382             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 383             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 384             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 385             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 386             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 387             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 388             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 389             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 390
 391             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velec            = _mm_and_ps(velec,cutoff_mask);
 395             velecsum         = _mm_add_ps(velecsum,velec);
 396
 397             fscal            = felec;
 398
 399             fscal            = _mm_and_ps(fscal,cutoff_mask);
 400
 401             /* Calculate temporary vectorial force */
 402             tx               = _mm_mul_ps(fscal,dx20);
 403             ty               = _mm_mul_ps(fscal,dy20);
 404             tz               = _mm_mul_ps(fscal,dz20);
 405
 406             /* Update vectorial force */
 407             fix2             = _mm_add_ps(fix2,tx);
 408             fiy2             = _mm_add_ps(fiy2,ty);
 409             fiz2             = _mm_add_ps(fiz2,tz);
 410
 411             fjx0             = _mm_add_ps(fjx0,tx);
 412             fjy0             = _mm_add_ps(fjy0,ty);
 413             fjz0             = _mm_add_ps(fjz0,tz);
 414
 415             }
 416
 417             /**************************
 418              * CALCULATE INTERACTIONS *
 419              **************************/
 420
 421             if (gmx_mm_any_lt(rsq30,rcutoff2))
 422             {
 423
 424             r30              = _mm_mul_ps(rsq30,rinv30);
 425
 426             /* Compute parameters for interactions between i and j atoms */
 427             qq30             = _mm_mul_ps(iq3,jq0);
 428
 429             /* EWALD ELECTROSTATICS */
 430
 431             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 432             ewrt             = _mm_mul_ps(r30,ewtabscale);
 433             ewitab           = _mm_cvttps_epi32(ewrt);
 434             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 435             ewitab           = _mm_slli_epi32(ewitab,2);
 436             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 437             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 438             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 439             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 440             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 441             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 442             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 443             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 444             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 445
 446             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 447
 448             /* Update potential sum for this i atom from the interaction with this j atom. */
 449             velec            = _mm_and_ps(velec,cutoff_mask);
 450             velecsum         = _mm_add_ps(velecsum,velec);
 451
 452             fscal            = felec;
 453
 454             fscal            = _mm_and_ps(fscal,cutoff_mask);
 455
 456             /* Calculate temporary vectorial force */
 457             tx               = _mm_mul_ps(fscal,dx30);
 458             ty               = _mm_mul_ps(fscal,dy30);
 459             tz               = _mm_mul_ps(fscal,dz30);
 460
 461             /* Update vectorial force */
 462             fix3             = _mm_add_ps(fix3,tx);
 463             fiy3             = _mm_add_ps(fiy3,ty);
 464             fiz3             = _mm_add_ps(fiz3,tz);
 465
 466             fjx0             = _mm_add_ps(fjx0,tx);
 467             fjy0             = _mm_add_ps(fjy0,ty);
 468             fjz0             = _mm_add_ps(fjz0,tz);
 469
 470             }
 471
 472             fjptrA             = f+j_coord_offsetA;
 473             fjptrB             = f+j_coord_offsetB;
 474             fjptrC             = f+j_coord_offsetC;
 475             fjptrD             = f+j_coord_offsetD;
 476
 477             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 478
 479             /* Inner loop uses 179 flops */
 480         }
 481
 482         if(jidx<j_index_end)
 483         {
 484
 485             /* Get j neighbor index, and coordinate index */
 486             jnrlistA         = jjnr[jidx];
 487             jnrlistB         = jjnr[jidx+1];
 488             jnrlistC         = jjnr[jidx+2];
 489             jnrlistD         = jjnr[jidx+3];
 490             /* Sign of each element will be negative for non-real atoms.
 491              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 492              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 493              */
 494             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 495             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 496             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 497             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 498             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 499             j_coord_offsetA  = DIM*jnrA;
 500             j_coord_offsetB  = DIM*jnrB;
 501             j_coord_offsetC  = DIM*jnrC;
 502             j_coord_offsetD  = DIM*jnrD;
 503
 504             /* load j atom coordinates */
 505             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 506                                               x+j_coord_offsetC,x+j_coord_offsetD,
 507                                               &jx0,&jy0,&jz0);
 508
 509             /* Calculate displacement vector */
 510             dx00             = _mm_sub_ps(ix0,jx0);
 511             dy00             = _mm_sub_ps(iy0,jy0);
 512             dz00             = _mm_sub_ps(iz0,jz0);
 513             dx10             = _mm_sub_ps(ix1,jx0);
 514             dy10             = _mm_sub_ps(iy1,jy0);
 515             dz10             = _mm_sub_ps(iz1,jz0);
 516             dx20             = _mm_sub_ps(ix2,jx0);
 517             dy20             = _mm_sub_ps(iy2,jy0);
 518             dz20             = _mm_sub_ps(iz2,jz0);
 519             dx30             = _mm_sub_ps(ix3,jx0);
 520             dy30             = _mm_sub_ps(iy3,jy0);
 521             dz30             = _mm_sub_ps(iz3,jz0);
 522
 523             /* Calculate squared distance and things based on it */
 524             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 525             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 526             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 527             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 528
 529             rinv10           = sse41_invsqrt_f(rsq10);
 530             rinv20           = sse41_invsqrt_f(rsq20);
 531             rinv30           = sse41_invsqrt_f(rsq30);
 532
 533             rinvsq00         = sse41_inv_f(rsq00);
 534             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 535             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 536             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 537
 538             /* Load parameters for j particles */
 539             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 540                                                               charge+jnrC+0,charge+jnrD+0);
 541             vdwjidx0A        = 2*vdwtype[jnrA+0];
 542             vdwjidx0B        = 2*vdwtype[jnrB+0];
 543             vdwjidx0C        = 2*vdwtype[jnrC+0];
 544             vdwjidx0D        = 2*vdwtype[jnrD+0];
 545
 546             fjx0             = _mm_setzero_ps();
 547             fjy0             = _mm_setzero_ps();
 548             fjz0             = _mm_setzero_ps();
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             if (gmx_mm_any_lt(rsq00,rcutoff2))
 555             {
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 559                                          vdwparam+vdwioffset0+vdwjidx0B,
 560                                          vdwparam+vdwioffset0+vdwjidx0C,
 561                                          vdwparam+vdwioffset0+vdwjidx0D,
 562                                          &c6_00,&c12_00);
 563
 564             /* LENNARD-JONES DISPERSION/REPULSION */
 565
 566             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 567             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 568             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 569             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 570                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 571             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 572
 573             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 574
 575             /* Update potential sum for this i atom from the interaction with this j atom. */
 576             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 577             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 578             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 579
 580             fscal            = fvdw;
 581
 582             fscal            = _mm_and_ps(fscal,cutoff_mask);
 583
 584             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 585
 586             /* Calculate temporary vectorial force */
 587             tx               = _mm_mul_ps(fscal,dx00);
 588             ty               = _mm_mul_ps(fscal,dy00);
 589             tz               = _mm_mul_ps(fscal,dz00);
 590
 591             /* Update vectorial force */
 592             fix0             = _mm_add_ps(fix0,tx);
 593             fiy0             = _mm_add_ps(fiy0,ty);
 594             fiz0             = _mm_add_ps(fiz0,tz);
 595
 596             fjx0             = _mm_add_ps(fjx0,tx);
 597             fjy0             = _mm_add_ps(fjy0,ty);
 598             fjz0             = _mm_add_ps(fjz0,tz);
 599
 600             }
 601
 602             /**************************
 603              * CALCULATE INTERACTIONS *
 604              **************************/
 605
 606             if (gmx_mm_any_lt(rsq10,rcutoff2))
 607             {
 608
 609             r10              = _mm_mul_ps(rsq10,rinv10);
 610             r10              = _mm_andnot_ps(dummy_mask,r10);
 611
 612             /* Compute parameters for interactions between i and j atoms */
 613             qq10             = _mm_mul_ps(iq1,jq0);
 614
 615             /* EWALD ELECTROSTATICS */
 616
 617             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 618             ewrt             = _mm_mul_ps(r10,ewtabscale);
 619             ewitab           = _mm_cvttps_epi32(ewrt);
 620             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 621             ewitab           = _mm_slli_epi32(ewitab,2);
 622             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 623             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 624             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 625             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 626             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 627             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 628             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 629             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 630             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 631
 632             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             velec            = _mm_and_ps(velec,cutoff_mask);
 636             velec            = _mm_andnot_ps(dummy_mask,velec);
 637             velecsum         = _mm_add_ps(velecsum,velec);
 638
 639             fscal            = felec;
 640
 641             fscal            = _mm_and_ps(fscal,cutoff_mask);
 642
 643             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm_mul_ps(fscal,dx10);
 647             ty               = _mm_mul_ps(fscal,dy10);
 648             tz               = _mm_mul_ps(fscal,dz10);
 649
 650             /* Update vectorial force */
 651             fix1             = _mm_add_ps(fix1,tx);
 652             fiy1             = _mm_add_ps(fiy1,ty);
 653             fiz1             = _mm_add_ps(fiz1,tz);
 654
 655             fjx0             = _mm_add_ps(fjx0,tx);
 656             fjy0             = _mm_add_ps(fjy0,ty);
 657             fjz0             = _mm_add_ps(fjz0,tz);
 658
 659             }
 660
 661             /**************************
 662              * CALCULATE INTERACTIONS *
 663              **************************/
 664
 665             if (gmx_mm_any_lt(rsq20,rcutoff2))
 666             {
 667
 668             r20              = _mm_mul_ps(rsq20,rinv20);
 669             r20              = _mm_andnot_ps(dummy_mask,r20);
 670
 671             /* Compute parameters for interactions between i and j atoms */
 672             qq20             = _mm_mul_ps(iq2,jq0);
 673
 674             /* EWALD ELECTROSTATICS */
 675
 676             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 677             ewrt             = _mm_mul_ps(r20,ewtabscale);
 678             ewitab           = _mm_cvttps_epi32(ewrt);
 679             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 680             ewitab           = _mm_slli_epi32(ewitab,2);
 681             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 682             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 683             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 684             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 685             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 686             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 687             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 688             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 689             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 690
 691             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 692
 693             /* Update potential sum for this i atom from the interaction with this j atom. */
 694             velec            = _mm_and_ps(velec,cutoff_mask);
 695             velec            = _mm_andnot_ps(dummy_mask,velec);
 696             velecsum         = _mm_add_ps(velecsum,velec);
 697
 698             fscal            = felec;
 699
 700             fscal            = _mm_and_ps(fscal,cutoff_mask);
 701
 702             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 703
 704             /* Calculate temporary vectorial force */
 705             tx               = _mm_mul_ps(fscal,dx20);
 706             ty               = _mm_mul_ps(fscal,dy20);
 707             tz               = _mm_mul_ps(fscal,dz20);
 708
 709             /* Update vectorial force */
 710             fix2             = _mm_add_ps(fix2,tx);
 711             fiy2             = _mm_add_ps(fiy2,ty);
 712             fiz2             = _mm_add_ps(fiz2,tz);
 713
 714             fjx0             = _mm_add_ps(fjx0,tx);
 715             fjy0             = _mm_add_ps(fjy0,ty);
 716             fjz0             = _mm_add_ps(fjz0,tz);
 717
 718             }
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             if (gmx_mm_any_lt(rsq30,rcutoff2))
 725             {
 726
 727             r30              = _mm_mul_ps(rsq30,rinv30);
 728             r30              = _mm_andnot_ps(dummy_mask,r30);
 729
 730             /* Compute parameters for interactions between i and j atoms */
 731             qq30             = _mm_mul_ps(iq3,jq0);
 732
 733             /* EWALD ELECTROSTATICS */
 734
 735             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 736             ewrt             = _mm_mul_ps(r30,ewtabscale);
 737             ewitab           = _mm_cvttps_epi32(ewrt);
 738             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 739             ewitab           = _mm_slli_epi32(ewitab,2);
 740             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 741             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 742             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 743             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 744             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 745             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 746             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 747             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 748             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 749
 750             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 751
 752             /* Update potential sum for this i atom from the interaction with this j atom. */
 753             velec            = _mm_and_ps(velec,cutoff_mask);
 754             velec            = _mm_andnot_ps(dummy_mask,velec);
 755             velecsum         = _mm_add_ps(velecsum,velec);
 756
 757             fscal            = felec;
 758
 759             fscal            = _mm_and_ps(fscal,cutoff_mask);
 760
 761             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 762
 763             /* Calculate temporary vectorial force */
 764             tx               = _mm_mul_ps(fscal,dx30);
 765             ty               = _mm_mul_ps(fscal,dy30);
 766             tz               = _mm_mul_ps(fscal,dz30);
 767
 768             /* Update vectorial force */
 769             fix3             = _mm_add_ps(fix3,tx);
 770             fiy3             = _mm_add_ps(fiy3,ty);
 771             fiz3             = _mm_add_ps(fiz3,tz);
 772
 773             fjx0             = _mm_add_ps(fjx0,tx);
 774             fjy0             = _mm_add_ps(fjy0,ty);
 775             fjz0             = _mm_add_ps(fjz0,tz);
 776
 777             }
 778
 779             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 780             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 781             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 782             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 783
 784             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 785
 786             /* Inner loop uses 182 flops */
 787         }
 788
 789         /* End of innermost loop */
 790
 791         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 792                                               f+i_coord_offset,fshift+i_shift_offset);
 793
 794         ggid                        = gid[iidx];
 795         /* Update potential energies */
 796         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 797         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 798
 799         /* Increment number of inner iterations */
 800         inneriter                  += j_index_end - j_index_start;
 801
 802         /* Outer loop uses 26 flops */
 803     }
 804
 805     /* Increment number of outer iterations */
 806     outeriter        += nri;
 807
 808     /* Update outer/inner flops */
 809
 810     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*182);
 811 }
 812 /*
 813  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse4_1_single
 814  * Electrostatics interaction: Ewald
 815  * VdW interaction:            LennardJones
 816  * Geometry:                   Water4-Particle
 817  * Calculate force/pot:        Force
 818  */
 819 void
 820 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse4_1_single
 821                     (t_nblist                    * gmx_restrict       nlist,
 822                      rvec                        * gmx_restrict          xx,
 823                      rvec                        * gmx_restrict          ff,
 824                      struct t_forcerec           * gmx_restrict          fr,
 825                      t_mdatoms                   * gmx_restrict     mdatoms,
 826                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 827                      t_nrnb                      * gmx_restrict        nrnb)
 828 {
 829     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 830      * just 0 for non-waters.
 831      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 832      * jnr indices corresponding to data put in the four positions in the SIMD register.
 833      */
 834     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 835     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 836     int              jnrA,jnrB,jnrC,jnrD;
 837     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 838     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 839     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 840     real             rcutoff_scalar;
 841     real             *shiftvec,*fshift,*x,*f;
 842     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 843     real             scratch[4*DIM];
 844     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 845     int              vdwioffset0;
 846     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 847     int              vdwioffset1;
 848     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 849     int              vdwioffset2;
 850     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 851     int              vdwioffset3;
 852     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 853     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 854     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 855     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 856     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 857     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 858     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 859     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 860     real             *charge;
 861     int              nvdwtype;
 862     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 863     int              *vdwtype;
 864     real             *vdwparam;
 865     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 866     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 867     __m128i          ewitab;
 868     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 869     real             *ewtab;
 870     __m128           dummy_mask,cutoff_mask;
 871     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 872     __m128           one     = _mm_set1_ps(1.0);
 873     __m128           two     = _mm_set1_ps(2.0);
 874     x                = xx[0];
 875     f                = ff[0];
 876
 877     nri              = nlist->nri;
 878     iinr             = nlist->iinr;
 879     jindex           = nlist->jindex;
 880     jjnr             = nlist->jjnr;
 881     shiftidx         = nlist->shift;
 882     gid              = nlist->gid;
 883     shiftvec         = fr->shift_vec[0];
 884     fshift           = fr->fshift[0];
 885     facel            = _mm_set1_ps(fr->ic->epsfac);
 886     charge           = mdatoms->chargeA;
 887     nvdwtype         = fr->ntype;
 888     vdwparam         = fr->nbfp;
 889     vdwtype          = mdatoms->typeA;
 890
 891     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 892     ewtab            = fr->ic->tabq_coul_F;
 893     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 894     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 895
 896     /* Setup water-specific parameters */
 897     inr              = nlist->iinr[0];
 898     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 899     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 900     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 901     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 902
 903     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 904     rcutoff_scalar   = fr->ic->rcoulomb;
 905     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 906     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 907
 908     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 909     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 910
 911     /* Avoid stupid compiler warnings */
 912     jnrA = jnrB = jnrC = jnrD = 0;
 913     j_coord_offsetA = 0;
 914     j_coord_offsetB = 0;
 915     j_coord_offsetC = 0;
 916     j_coord_offsetD = 0;
 917
 918     outeriter        = 0;
 919     inneriter        = 0;
 920
 921     for(iidx=0;iidx<4*DIM;iidx++)
 922     {
 923         scratch[iidx] = 0.0;
 924     }
 925
 926     /* Start outer loop over neighborlists */
 927     for(iidx=0; iidx<nri; iidx++)
 928     {
 929         /* Load shift vector for this list */
 930         i_shift_offset   = DIM*shiftidx[iidx];
 931
 932         /* Load limits for loop over neighbors */
 933         j_index_start    = jindex[iidx];
 934         j_index_end      = jindex[iidx+1];
 935
 936         /* Get outer coordinate index */
 937         inr              = iinr[iidx];
 938         i_coord_offset   = DIM*inr;
 939
 940         /* Load i particle coords and add shift vector */
 941         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 942                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 943
 944         fix0             = _mm_setzero_ps();
 945         fiy0             = _mm_setzero_ps();
 946         fiz0             = _mm_setzero_ps();
 947         fix1             = _mm_setzero_ps();
 948         fiy1             = _mm_setzero_ps();
 949         fiz1             = _mm_setzero_ps();
 950         fix2             = _mm_setzero_ps();
 951         fiy2             = _mm_setzero_ps();
 952         fiz2             = _mm_setzero_ps();
 953         fix3             = _mm_setzero_ps();
 954         fiy3             = _mm_setzero_ps();
 955         fiz3             = _mm_setzero_ps();
 956
 957         /* Start inner kernel loop */
 958         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 959         {
 960
 961             /* Get j neighbor index, and coordinate index */
 962             jnrA             = jjnr[jidx];
 963             jnrB             = jjnr[jidx+1];
 964             jnrC             = jjnr[jidx+2];
 965             jnrD             = jjnr[jidx+3];
 966             j_coord_offsetA  = DIM*jnrA;
 967             j_coord_offsetB  = DIM*jnrB;
 968             j_coord_offsetC  = DIM*jnrC;
 969             j_coord_offsetD  = DIM*jnrD;
 970
 971             /* load j atom coordinates */
 972             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 973                                               x+j_coord_offsetC,x+j_coord_offsetD,
 974                                               &jx0,&jy0,&jz0);
 975
 976             /* Calculate displacement vector */
 977             dx00             = _mm_sub_ps(ix0,jx0);
 978             dy00             = _mm_sub_ps(iy0,jy0);
 979             dz00             = _mm_sub_ps(iz0,jz0);
 980             dx10             = _mm_sub_ps(ix1,jx0);
 981             dy10             = _mm_sub_ps(iy1,jy0);
 982             dz10             = _mm_sub_ps(iz1,jz0);
 983             dx20             = _mm_sub_ps(ix2,jx0);
 984             dy20             = _mm_sub_ps(iy2,jy0);
 985             dz20             = _mm_sub_ps(iz2,jz0);
 986             dx30             = _mm_sub_ps(ix3,jx0);
 987             dy30             = _mm_sub_ps(iy3,jy0);
 988             dz30             = _mm_sub_ps(iz3,jz0);
 989
 990             /* Calculate squared distance and things based on it */
 991             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 992             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 993             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 994             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 995
 996             rinv10           = sse41_invsqrt_f(rsq10);
 997             rinv20           = sse41_invsqrt_f(rsq20);
 998             rinv30           = sse41_invsqrt_f(rsq30);
 999
1000             rinvsq00         = sse41_inv_f(rsq00);
1001             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1002             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1003             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1004
1005             /* Load parameters for j particles */
1006             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1007                                                               charge+jnrC+0,charge+jnrD+0);
1008             vdwjidx0A        = 2*vdwtype[jnrA+0];
1009             vdwjidx0B        = 2*vdwtype[jnrB+0];
1010             vdwjidx0C        = 2*vdwtype[jnrC+0];
1011             vdwjidx0D        = 2*vdwtype[jnrD+0];
1012
1013             fjx0             = _mm_setzero_ps();
1014             fjy0             = _mm_setzero_ps();
1015             fjz0             = _mm_setzero_ps();
1016
1017             /**************************
1018              * CALCULATE INTERACTIONS *
1019              **************************/
1020
1021             if (gmx_mm_any_lt(rsq00,rcutoff2))
1022             {
1023
1024             /* Compute parameters for interactions between i and j atoms */
1025             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1026                                          vdwparam+vdwioffset0+vdwjidx0B,
1027                                          vdwparam+vdwioffset0+vdwjidx0C,
1028                                          vdwparam+vdwioffset0+vdwjidx0D,
1029                                          &c6_00,&c12_00);
1030
1031             /* LENNARD-JONES DISPERSION/REPULSION */
1032
1033             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1034             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1035
1036             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1037
1038             fscal            = fvdw;
1039
1040             fscal            = _mm_and_ps(fscal,cutoff_mask);
1041
1042             /* Calculate temporary vectorial force */
1043             tx               = _mm_mul_ps(fscal,dx00);
1044             ty               = _mm_mul_ps(fscal,dy00);
1045             tz               = _mm_mul_ps(fscal,dz00);
1046
1047             /* Update vectorial force */
1048             fix0             = _mm_add_ps(fix0,tx);
1049             fiy0             = _mm_add_ps(fiy0,ty);
1050             fiz0             = _mm_add_ps(fiz0,tz);
1051
1052             fjx0             = _mm_add_ps(fjx0,tx);
1053             fjy0             = _mm_add_ps(fjy0,ty);
1054             fjz0             = _mm_add_ps(fjz0,tz);
1055
1056             }
1057
1058             /**************************
1059              * CALCULATE INTERACTIONS *
1060              **************************/
1061
1062             if (gmx_mm_any_lt(rsq10,rcutoff2))
1063             {
1064
1065             r10              = _mm_mul_ps(rsq10,rinv10);
1066
1067             /* Compute parameters for interactions between i and j atoms */
1068             qq10             = _mm_mul_ps(iq1,jq0);
1069
1070             /* EWALD ELECTROSTATICS */
1071
1072             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1073             ewrt             = _mm_mul_ps(r10,ewtabscale);
1074             ewitab           = _mm_cvttps_epi32(ewrt);
1075             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1076             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1077                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1078                                          &ewtabF,&ewtabFn);
1079             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1080             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1081
1082             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1083
1084             fscal            = felec;
1085
1086             fscal            = _mm_and_ps(fscal,cutoff_mask);
1087
1088             /* Calculate temporary vectorial force */
1089             tx               = _mm_mul_ps(fscal,dx10);
1090             ty               = _mm_mul_ps(fscal,dy10);
1091             tz               = _mm_mul_ps(fscal,dz10);
1092
1093             /* Update vectorial force */
1094             fix1             = _mm_add_ps(fix1,tx);
1095             fiy1             = _mm_add_ps(fiy1,ty);
1096             fiz1             = _mm_add_ps(fiz1,tz);
1097
1098             fjx0             = _mm_add_ps(fjx0,tx);
1099             fjy0             = _mm_add_ps(fjy0,ty);
1100             fjz0             = _mm_add_ps(fjz0,tz);
1101
1102             }
1103
1104             /**************************
1105              * CALCULATE INTERACTIONS *
1106              **************************/
1107
1108             if (gmx_mm_any_lt(rsq20,rcutoff2))
1109             {
1110
1111             r20              = _mm_mul_ps(rsq20,rinv20);
1112
1113             /* Compute parameters for interactions between i and j atoms */
1114             qq20             = _mm_mul_ps(iq2,jq0);
1115
1116             /* EWALD ELECTROSTATICS */
1117
1118             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1119             ewrt             = _mm_mul_ps(r20,ewtabscale);
1120             ewitab           = _mm_cvttps_epi32(ewrt);
1121             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1122             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1123                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1124                                          &ewtabF,&ewtabFn);
1125             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1126             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1127
1128             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1129
1130             fscal            = felec;
1131
1132             fscal            = _mm_and_ps(fscal,cutoff_mask);
1133
1134             /* Calculate temporary vectorial force */
1135             tx               = _mm_mul_ps(fscal,dx20);
1136             ty               = _mm_mul_ps(fscal,dy20);
1137             tz               = _mm_mul_ps(fscal,dz20);
1138
1139             /* Update vectorial force */
1140             fix2             = _mm_add_ps(fix2,tx);
1141             fiy2             = _mm_add_ps(fiy2,ty);
1142             fiz2             = _mm_add_ps(fiz2,tz);
1143
1144             fjx0             = _mm_add_ps(fjx0,tx);
1145             fjy0             = _mm_add_ps(fjy0,ty);
1146             fjz0             = _mm_add_ps(fjz0,tz);
1147
1148             }
1149
1150             /**************************
1151              * CALCULATE INTERACTIONS *
1152              **************************/
1153
1154             if (gmx_mm_any_lt(rsq30,rcutoff2))
1155             {
1156
1157             r30              = _mm_mul_ps(rsq30,rinv30);
1158
1159             /* Compute parameters for interactions between i and j atoms */
1160             qq30             = _mm_mul_ps(iq3,jq0);
1161
1162             /* EWALD ELECTROSTATICS */
1163
1164             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1165             ewrt             = _mm_mul_ps(r30,ewtabscale);
1166             ewitab           = _mm_cvttps_epi32(ewrt);
1167             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1168             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1169                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1170                                          &ewtabF,&ewtabFn);
1171             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1172             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1173
1174             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1175
1176             fscal            = felec;
1177
1178             fscal            = _mm_and_ps(fscal,cutoff_mask);
1179
1180             /* Calculate temporary vectorial force */
1181             tx               = _mm_mul_ps(fscal,dx30);
1182             ty               = _mm_mul_ps(fscal,dy30);
1183             tz               = _mm_mul_ps(fscal,dz30);
1184
1185             /* Update vectorial force */
1186             fix3             = _mm_add_ps(fix3,tx);
1187             fiy3             = _mm_add_ps(fiy3,ty);
1188             fiz3             = _mm_add_ps(fiz3,tz);
1189
1190             fjx0             = _mm_add_ps(fjx0,tx);
1191             fjy0             = _mm_add_ps(fjy0,ty);
1192             fjz0             = _mm_add_ps(fjz0,tz);
1193
1194             }
1195
1196             fjptrA             = f+j_coord_offsetA;
1197             fjptrB             = f+j_coord_offsetB;
1198             fjptrC             = f+j_coord_offsetC;
1199             fjptrD             = f+j_coord_offsetD;
1200
1201             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1202
1203             /* Inner loop uses 147 flops */
1204         }
1205
1206         if(jidx<j_index_end)
1207         {
1208
1209             /* Get j neighbor index, and coordinate index */
1210             jnrlistA         = jjnr[jidx];
1211             jnrlistB         = jjnr[jidx+1];
1212             jnrlistC         = jjnr[jidx+2];
1213             jnrlistD         = jjnr[jidx+3];
1214             /* Sign of each element will be negative for non-real atoms.
1215              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1216              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1217              */
1218             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1219             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1220             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1221             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1222             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1223             j_coord_offsetA  = DIM*jnrA;
1224             j_coord_offsetB  = DIM*jnrB;
1225             j_coord_offsetC  = DIM*jnrC;
1226             j_coord_offsetD  = DIM*jnrD;
1227
1228             /* load j atom coordinates */
1229             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1230                                               x+j_coord_offsetC,x+j_coord_offsetD,
1231                                               &jx0,&jy0,&jz0);
1232
1233             /* Calculate displacement vector */
1234             dx00             = _mm_sub_ps(ix0,jx0);
1235             dy00             = _mm_sub_ps(iy0,jy0);
1236             dz00             = _mm_sub_ps(iz0,jz0);
1237             dx10             = _mm_sub_ps(ix1,jx0);
1238             dy10             = _mm_sub_ps(iy1,jy0);
1239             dz10             = _mm_sub_ps(iz1,jz0);
1240             dx20             = _mm_sub_ps(ix2,jx0);
1241             dy20             = _mm_sub_ps(iy2,jy0);
1242             dz20             = _mm_sub_ps(iz2,jz0);
1243             dx30             = _mm_sub_ps(ix3,jx0);
1244             dy30             = _mm_sub_ps(iy3,jy0);
1245             dz30             = _mm_sub_ps(iz3,jz0);
1246
1247             /* Calculate squared distance and things based on it */
1248             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1249             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1250             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1251             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1252
1253             rinv10           = sse41_invsqrt_f(rsq10);
1254             rinv20           = sse41_invsqrt_f(rsq20);
1255             rinv30           = sse41_invsqrt_f(rsq30);
1256
1257             rinvsq00         = sse41_inv_f(rsq00);
1258             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1259             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1260             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1261
1262             /* Load parameters for j particles */
1263             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1264                                                               charge+jnrC+0,charge+jnrD+0);
1265             vdwjidx0A        = 2*vdwtype[jnrA+0];
1266             vdwjidx0B        = 2*vdwtype[jnrB+0];
1267             vdwjidx0C        = 2*vdwtype[jnrC+0];
1268             vdwjidx0D        = 2*vdwtype[jnrD+0];
1269
1270             fjx0             = _mm_setzero_ps();
1271             fjy0             = _mm_setzero_ps();
1272             fjz0             = _mm_setzero_ps();
1273
1274             /**************************
1275              * CALCULATE INTERACTIONS *
1276              **************************/
1277
1278             if (gmx_mm_any_lt(rsq00,rcutoff2))
1279             {
1280
1281             /* Compute parameters for interactions between i and j atoms */
1282             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1283                                          vdwparam+vdwioffset0+vdwjidx0B,
1284                                          vdwparam+vdwioffset0+vdwjidx0C,
1285                                          vdwparam+vdwioffset0+vdwjidx0D,
1286                                          &c6_00,&c12_00);
1287
1288             /* LENNARD-JONES DISPERSION/REPULSION */
1289
1290             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1291             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1292
1293             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1294
1295             fscal            = fvdw;
1296
1297             fscal            = _mm_and_ps(fscal,cutoff_mask);
1298
1299             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1300
1301             /* Calculate temporary vectorial force */
1302             tx               = _mm_mul_ps(fscal,dx00);
1303             ty               = _mm_mul_ps(fscal,dy00);
1304             tz               = _mm_mul_ps(fscal,dz00);
1305
1306             /* Update vectorial force */
1307             fix0             = _mm_add_ps(fix0,tx);
1308             fiy0             = _mm_add_ps(fiy0,ty);
1309             fiz0             = _mm_add_ps(fiz0,tz);
1310
1311             fjx0             = _mm_add_ps(fjx0,tx);
1312             fjy0             = _mm_add_ps(fjy0,ty);
1313             fjz0             = _mm_add_ps(fjz0,tz);
1314
1315             }
1316
1317             /**************************
1318              * CALCULATE INTERACTIONS *
1319              **************************/
1320
1321             if (gmx_mm_any_lt(rsq10,rcutoff2))
1322             {
1323
1324             r10              = _mm_mul_ps(rsq10,rinv10);
1325             r10              = _mm_andnot_ps(dummy_mask,r10);
1326
1327             /* Compute parameters for interactions between i and j atoms */
1328             qq10             = _mm_mul_ps(iq1,jq0);
1329
1330             /* EWALD ELECTROSTATICS */
1331
1332             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1333             ewrt             = _mm_mul_ps(r10,ewtabscale);
1334             ewitab           = _mm_cvttps_epi32(ewrt);
1335             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1336             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1337                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1338                                          &ewtabF,&ewtabFn);
1339             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1340             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1341
1342             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1343
1344             fscal            = felec;
1345
1346             fscal            = _mm_and_ps(fscal,cutoff_mask);
1347
1348             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1349
1350             /* Calculate temporary vectorial force */
1351             tx               = _mm_mul_ps(fscal,dx10);
1352             ty               = _mm_mul_ps(fscal,dy10);
1353             tz               = _mm_mul_ps(fscal,dz10);
1354
1355             /* Update vectorial force */
1356             fix1             = _mm_add_ps(fix1,tx);
1357             fiy1             = _mm_add_ps(fiy1,ty);
1358             fiz1             = _mm_add_ps(fiz1,tz);
1359
1360             fjx0             = _mm_add_ps(fjx0,tx);
1361             fjy0             = _mm_add_ps(fjy0,ty);
1362             fjz0             = _mm_add_ps(fjz0,tz);
1363
1364             }
1365
1366             /**************************
1367              * CALCULATE INTERACTIONS *
1368              **************************/
1369
1370             if (gmx_mm_any_lt(rsq20,rcutoff2))
1371             {
1372
1373             r20              = _mm_mul_ps(rsq20,rinv20);
1374             r20              = _mm_andnot_ps(dummy_mask,r20);
1375
1376             /* Compute parameters for interactions between i and j atoms */
1377             qq20             = _mm_mul_ps(iq2,jq0);
1378
1379             /* EWALD ELECTROSTATICS */
1380
1381             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1382             ewrt             = _mm_mul_ps(r20,ewtabscale);
1383             ewitab           = _mm_cvttps_epi32(ewrt);
1384             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1385             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1386                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1387                                          &ewtabF,&ewtabFn);
1388             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1389             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1390
1391             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1392
1393             fscal            = felec;
1394
1395             fscal            = _mm_and_ps(fscal,cutoff_mask);
1396
1397             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1398
1399             /* Calculate temporary vectorial force */
1400             tx               = _mm_mul_ps(fscal,dx20);
1401             ty               = _mm_mul_ps(fscal,dy20);
1402             tz               = _mm_mul_ps(fscal,dz20);
1403
1404             /* Update vectorial force */
1405             fix2             = _mm_add_ps(fix2,tx);
1406             fiy2             = _mm_add_ps(fiy2,ty);
1407             fiz2             = _mm_add_ps(fiz2,tz);
1408
1409             fjx0             = _mm_add_ps(fjx0,tx);
1410             fjy0             = _mm_add_ps(fjy0,ty);
1411             fjz0             = _mm_add_ps(fjz0,tz);
1412
1413             }
1414
1415             /**************************
1416              * CALCULATE INTERACTIONS *
1417              **************************/
1418
1419             if (gmx_mm_any_lt(rsq30,rcutoff2))
1420             {
1421
1422             r30              = _mm_mul_ps(rsq30,rinv30);
1423             r30              = _mm_andnot_ps(dummy_mask,r30);
1424
1425             /* Compute parameters for interactions between i and j atoms */
1426             qq30             = _mm_mul_ps(iq3,jq0);
1427
1428             /* EWALD ELECTROSTATICS */
1429
1430             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1431             ewrt             = _mm_mul_ps(r30,ewtabscale);
1432             ewitab           = _mm_cvttps_epi32(ewrt);
1433             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1434             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1435                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1436                                          &ewtabF,&ewtabFn);
1437             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1438             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1439
1440             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1441
1442             fscal            = felec;
1443
1444             fscal            = _mm_and_ps(fscal,cutoff_mask);
1445
1446             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1447
1448             /* Calculate temporary vectorial force */
1449             tx               = _mm_mul_ps(fscal,dx30);
1450             ty               = _mm_mul_ps(fscal,dy30);
1451             tz               = _mm_mul_ps(fscal,dz30);
1452
1453             /* Update vectorial force */
1454             fix3             = _mm_add_ps(fix3,tx);
1455             fiy3             = _mm_add_ps(fiy3,ty);
1456             fiz3             = _mm_add_ps(fiz3,tz);
1457
1458             fjx0             = _mm_add_ps(fjx0,tx);
1459             fjy0             = _mm_add_ps(fjy0,ty);
1460             fjz0             = _mm_add_ps(fjz0,tz);
1461
1462             }
1463
1464             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1465             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1466             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1467             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1468
1469             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1470
1471             /* Inner loop uses 150 flops */
1472         }
1473
1474         /* End of innermost loop */
1475
1476         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1477                                               f+i_coord_offset,fshift+i_shift_offset);
1478
1479         /* Increment number of inner iterations */
1480         inneriter                  += j_index_end - j_index_start;
1481
1482         /* Outer loop uses 24 flops */
1483     }
1484
1485     /* Increment number of outer iterations */
1486     outeriter        += nri;
1487
1488     /* Update outer/inner flops */
1489
1490     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1491 }