src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_sse2_single.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 sse2_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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse2_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  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     __m128           c6grid_00;
 105     __m128           c6grid_10;
 106     __m128           c6grid_20;
 107     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 108     real             *vdwgridparam;
 109     __m128           one_half = _mm_set1_ps(0.5);
 110     __m128           minus_one = _mm_set1_ps(-1.0);
 111     __m128i          ewitab;
 112     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     real             *ewtab;
 114     __m128           dummy_mask,cutoff_mask;
 115     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 116     __m128           one     = _mm_set1_ps(1.0);
 117     __m128           two     = _mm_set1_ps(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm_set1_ps(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134     vdwgridparam     = fr->ljpme_c6grid;
 135     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 136     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 137     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 138
 139     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 140     ewtab            = fr->ic->tabq_coul_FDV0;
 141     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 142     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 147     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 148     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 149     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 152     rcutoff_scalar   = fr->rcoulomb;
 153     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 154     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 155
 156     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 157     rvdw             = _mm_set1_ps(fr->rvdw);
 158
 159     /* Avoid stupid compiler warnings */
 160     jnrA = jnrB = jnrC = jnrD = 0;
 161     j_coord_offsetA = 0;
 162     j_coord_offsetB = 0;
 163     j_coord_offsetC = 0;
 164     j_coord_offsetD = 0;
 165
 166     outeriter        = 0;
 167     inneriter        = 0;
 168
 169     for(iidx=0;iidx<4*DIM;iidx++)
 170     {
 171         scratch[iidx] = 0.0;
 172     }
 173
 174     /* Start outer loop over neighborlists */
 175     for(iidx=0; iidx<nri; iidx++)
 176     {
 177         /* Load shift vector for this list */
 178         i_shift_offset   = DIM*shiftidx[iidx];
 179
 180         /* Load limits for loop over neighbors */
 181         j_index_start    = jindex[iidx];
 182         j_index_end      = jindex[iidx+1];
 183
 184         /* Get outer coordinate index */
 185         inr              = iinr[iidx];
 186         i_coord_offset   = DIM*inr;
 187
 188         /* Load i particle coords and add shift vector */
 189         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 190                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 191
 192         fix0             = _mm_setzero_ps();
 193         fiy0             = _mm_setzero_ps();
 194         fiz0             = _mm_setzero_ps();
 195         fix1             = _mm_setzero_ps();
 196         fiy1             = _mm_setzero_ps();
 197         fiz1             = _mm_setzero_ps();
 198         fix2             = _mm_setzero_ps();
 199         fiy2             = _mm_setzero_ps();
 200         fiz2             = _mm_setzero_ps();
 201
 202         /* Reset potential sums */
 203         velecsum         = _mm_setzero_ps();
 204         vvdwsum          = _mm_setzero_ps();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             jnrC             = jjnr[jidx+2];
 214             jnrD             = jjnr[jidx+3];
 215             j_coord_offsetA  = DIM*jnrA;
 216             j_coord_offsetB  = DIM*jnrB;
 217             j_coord_offsetC  = DIM*jnrC;
 218             j_coord_offsetD  = DIM*jnrD;
 219
 220             /* load j atom coordinates */
 221             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                               x+j_coord_offsetC,x+j_coord_offsetD,
 223                                               &jx0,&jy0,&jz0);
 224
 225             /* Calculate displacement vector */
 226             dx00             = _mm_sub_ps(ix0,jx0);
 227             dy00             = _mm_sub_ps(iy0,jy0);
 228             dz00             = _mm_sub_ps(iz0,jz0);
 229             dx10             = _mm_sub_ps(ix1,jx0);
 230             dy10             = _mm_sub_ps(iy1,jy0);
 231             dz10             = _mm_sub_ps(iz1,jz0);
 232             dx20             = _mm_sub_ps(ix2,jx0);
 233             dy20             = _mm_sub_ps(iy2,jy0);
 234             dz20             = _mm_sub_ps(iz2,jz0);
 235
 236             /* Calculate squared distance and things based on it */
 237             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 238             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 239             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 240
 241             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 242             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 243             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 244
 245             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 246             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 247             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 248
 249             /* Load parameters for j particles */
 250             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 251                                                               charge+jnrC+0,charge+jnrD+0);
 252             vdwjidx0A        = 2*vdwtype[jnrA+0];
 253             vdwjidx0B        = 2*vdwtype[jnrB+0];
 254             vdwjidx0C        = 2*vdwtype[jnrC+0];
 255             vdwjidx0D        = 2*vdwtype[jnrD+0];
 256
 257             fjx0             = _mm_setzero_ps();
 258             fjy0             = _mm_setzero_ps();
 259             fjz0             = _mm_setzero_ps();
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             if (gmx_mm_any_lt(rsq00,rcutoff2))
 266             {
 267
 268             r00              = _mm_mul_ps(rsq00,rinv00);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq00             = _mm_mul_ps(iq0,jq0);
 272             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 273                                          vdwparam+vdwioffset0+vdwjidx0B,
 274                                          vdwparam+vdwioffset0+vdwjidx0C,
 275                                          vdwparam+vdwioffset0+vdwjidx0D,
 276                                          &c6_00,&c12_00);
 277             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 278                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 279                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 280                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 281
 282             /* EWALD ELECTROSTATICS */
 283
 284             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 285             ewrt             = _mm_mul_ps(r00,ewtabscale);
 286             ewitab           = _mm_cvttps_epi32(ewrt);
 287             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 288             ewitab           = _mm_slli_epi32(ewitab,2);
 289             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 290             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 291             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 292             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 293             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 294             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 295             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 296             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 297             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 298
 299             /* Analytical LJ-PME */
 300             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 301             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 302             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 303             exponent         = gmx_simd_exp_r(ewcljrsq);
 304             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 305             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 306             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 307             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 308             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 309             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) ,
 310                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 311             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 312             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 313
 314             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velec            = _mm_and_ps(velec,cutoff_mask);
 318             velecsum         = _mm_add_ps(velecsum,velec);
 319             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 320             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 321
 322             fscal            = _mm_add_ps(felec,fvdw);
 323
 324             fscal            = _mm_and_ps(fscal,cutoff_mask);
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm_mul_ps(fscal,dx00);
 328             ty               = _mm_mul_ps(fscal,dy00);
 329             tz               = _mm_mul_ps(fscal,dz00);
 330
 331             /* Update vectorial force */
 332             fix0             = _mm_add_ps(fix0,tx);
 333             fiy0             = _mm_add_ps(fiy0,ty);
 334             fiz0             = _mm_add_ps(fiz0,tz);
 335
 336             fjx0             = _mm_add_ps(fjx0,tx);
 337             fjy0             = _mm_add_ps(fjy0,ty);
 338             fjz0             = _mm_add_ps(fjz0,tz);
 339
 340             }
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             if (gmx_mm_any_lt(rsq10,rcutoff2))
 347             {
 348
 349             r10              = _mm_mul_ps(rsq10,rinv10);
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq10             = _mm_mul_ps(iq1,jq0);
 353
 354             /* EWALD ELECTROSTATICS */
 355
 356             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 357             ewrt             = _mm_mul_ps(r10,ewtabscale);
 358             ewitab           = _mm_cvttps_epi32(ewrt);
 359             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 360             ewitab           = _mm_slli_epi32(ewitab,2);
 361             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 362             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 363             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 364             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 365             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 366             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 367             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 368             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 369             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 370
 371             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velec            = _mm_and_ps(velec,cutoff_mask);
 375             velecsum         = _mm_add_ps(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             fscal            = _mm_and_ps(fscal,cutoff_mask);
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm_mul_ps(fscal,dx10);
 383             ty               = _mm_mul_ps(fscal,dy10);
 384             tz               = _mm_mul_ps(fscal,dz10);
 385
 386             /* Update vectorial force */
 387             fix1             = _mm_add_ps(fix1,tx);
 388             fiy1             = _mm_add_ps(fiy1,ty);
 389             fiz1             = _mm_add_ps(fiz1,tz);
 390
 391             fjx0             = _mm_add_ps(fjx0,tx);
 392             fjy0             = _mm_add_ps(fjy0,ty);
 393             fjz0             = _mm_add_ps(fjz0,tz);
 394
 395             }
 396
 397             /**************************
 398              * CALCULATE INTERACTIONS *
 399              **************************/
 400
 401             if (gmx_mm_any_lt(rsq20,rcutoff2))
 402             {
 403
 404             r20              = _mm_mul_ps(rsq20,rinv20);
 405
 406             /* Compute parameters for interactions between i and j atoms */
 407             qq20             = _mm_mul_ps(iq2,jq0);
 408
 409             /* EWALD ELECTROSTATICS */
 410
 411             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 412             ewrt             = _mm_mul_ps(r20,ewtabscale);
 413             ewitab           = _mm_cvttps_epi32(ewrt);
 414             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 415             ewitab           = _mm_slli_epi32(ewitab,2);
 416             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 417             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 418             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 419             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 420             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 421             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 422             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 423             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 424             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 425
 426             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velec            = _mm_and_ps(velec,cutoff_mask);
 430             velecsum         = _mm_add_ps(velecsum,velec);
 431
 432             fscal            = felec;
 433
 434             fscal            = _mm_and_ps(fscal,cutoff_mask);
 435
 436             /* Calculate temporary vectorial force */
 437             tx               = _mm_mul_ps(fscal,dx20);
 438             ty               = _mm_mul_ps(fscal,dy20);
 439             tz               = _mm_mul_ps(fscal,dz20);
 440
 441             /* Update vectorial force */
 442             fix2             = _mm_add_ps(fix2,tx);
 443             fiy2             = _mm_add_ps(fiy2,ty);
 444             fiz2             = _mm_add_ps(fiz2,tz);
 445
 446             fjx0             = _mm_add_ps(fjx0,tx);
 447             fjy0             = _mm_add_ps(fjy0,ty);
 448             fjz0             = _mm_add_ps(fjz0,tz);
 449
 450             }
 451
 452             fjptrA             = f+j_coord_offsetA;
 453             fjptrB             = f+j_coord_offsetB;
 454             fjptrC             = f+j_coord_offsetC;
 455             fjptrD             = f+j_coord_offsetD;
 456
 457             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 458
 459             /* Inner loop uses 174 flops */
 460         }
 461
 462         if(jidx<j_index_end)
 463         {
 464
 465             /* Get j neighbor index, and coordinate index */
 466             jnrlistA         = jjnr[jidx];
 467             jnrlistB         = jjnr[jidx+1];
 468             jnrlistC         = jjnr[jidx+2];
 469             jnrlistD         = jjnr[jidx+3];
 470             /* Sign of each element will be negative for non-real atoms.
 471              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 472              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 473              */
 474             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 475             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 476             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 477             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 478             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 479             j_coord_offsetA  = DIM*jnrA;
 480             j_coord_offsetB  = DIM*jnrB;
 481             j_coord_offsetC  = DIM*jnrC;
 482             j_coord_offsetD  = DIM*jnrD;
 483
 484             /* load j atom coordinates */
 485             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 486                                               x+j_coord_offsetC,x+j_coord_offsetD,
 487                                               &jx0,&jy0,&jz0);
 488
 489             /* Calculate displacement vector */
 490             dx00             = _mm_sub_ps(ix0,jx0);
 491             dy00             = _mm_sub_ps(iy0,jy0);
 492             dz00             = _mm_sub_ps(iz0,jz0);
 493             dx10             = _mm_sub_ps(ix1,jx0);
 494             dy10             = _mm_sub_ps(iy1,jy0);
 495             dz10             = _mm_sub_ps(iz1,jz0);
 496             dx20             = _mm_sub_ps(ix2,jx0);
 497             dy20             = _mm_sub_ps(iy2,jy0);
 498             dz20             = _mm_sub_ps(iz2,jz0);
 499
 500             /* Calculate squared distance and things based on it */
 501             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 502             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 503             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 504
 505             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 506             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 507             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 508
 509             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 510             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 511             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 512
 513             /* Load parameters for j particles */
 514             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 515                                                               charge+jnrC+0,charge+jnrD+0);
 516             vdwjidx0A        = 2*vdwtype[jnrA+0];
 517             vdwjidx0B        = 2*vdwtype[jnrB+0];
 518             vdwjidx0C        = 2*vdwtype[jnrC+0];
 519             vdwjidx0D        = 2*vdwtype[jnrD+0];
 520
 521             fjx0             = _mm_setzero_ps();
 522             fjy0             = _mm_setzero_ps();
 523             fjz0             = _mm_setzero_ps();
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             if (gmx_mm_any_lt(rsq00,rcutoff2))
 530             {
 531
 532             r00              = _mm_mul_ps(rsq00,rinv00);
 533             r00              = _mm_andnot_ps(dummy_mask,r00);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq00             = _mm_mul_ps(iq0,jq0);
 537             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 538                                          vdwparam+vdwioffset0+vdwjidx0B,
 539                                          vdwparam+vdwioffset0+vdwjidx0C,
 540                                          vdwparam+vdwioffset0+vdwjidx0D,
 541                                          &c6_00,&c12_00);
 542             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 543                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 544                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 545                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 546
 547             /* EWALD ELECTROSTATICS */
 548
 549             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 550             ewrt             = _mm_mul_ps(r00,ewtabscale);
 551             ewitab           = _mm_cvttps_epi32(ewrt);
 552             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 553             ewitab           = _mm_slli_epi32(ewitab,2);
 554             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 555             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 556             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 557             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 558             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 559             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 560             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 561             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 562             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 563
 564             /* Analytical LJ-PME */
 565             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 566             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 567             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 568             exponent         = gmx_simd_exp_r(ewcljrsq);
 569             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 570             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 571             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 572             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 573             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 574             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) ,
 575                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 576             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 577             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 578
 579             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm_and_ps(velec,cutoff_mask);
 583             velec            = _mm_andnot_ps(dummy_mask,velec);
 584             velecsum         = _mm_add_ps(velecsum,velec);
 585             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 586             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 587             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 588
 589             fscal            = _mm_add_ps(felec,fvdw);
 590
 591             fscal            = _mm_and_ps(fscal,cutoff_mask);
 592
 593             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm_mul_ps(fscal,dx00);
 597             ty               = _mm_mul_ps(fscal,dy00);
 598             tz               = _mm_mul_ps(fscal,dz00);
 599
 600             /* Update vectorial force */
 601             fix0             = _mm_add_ps(fix0,tx);
 602             fiy0             = _mm_add_ps(fiy0,ty);
 603             fiz0             = _mm_add_ps(fiz0,tz);
 604
 605             fjx0             = _mm_add_ps(fjx0,tx);
 606             fjy0             = _mm_add_ps(fjy0,ty);
 607             fjz0             = _mm_add_ps(fjz0,tz);
 608
 609             }
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             if (gmx_mm_any_lt(rsq10,rcutoff2))
 616             {
 617
 618             r10              = _mm_mul_ps(rsq10,rinv10);
 619             r10              = _mm_andnot_ps(dummy_mask,r10);
 620
 621             /* Compute parameters for interactions between i and j atoms */
 622             qq10             = _mm_mul_ps(iq1,jq0);
 623
 624             /* EWALD ELECTROSTATICS */
 625
 626             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 627             ewrt             = _mm_mul_ps(r10,ewtabscale);
 628             ewitab           = _mm_cvttps_epi32(ewrt);
 629             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 630             ewitab           = _mm_slli_epi32(ewitab,2);
 631             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 632             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 633             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 634             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 635             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 636             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 637             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 638             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 639             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 640
 641             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 642
 643             /* Update potential sum for this i atom from the interaction with this j atom. */
 644             velec            = _mm_and_ps(velec,cutoff_mask);
 645             velec            = _mm_andnot_ps(dummy_mask,velec);
 646             velecsum         = _mm_add_ps(velecsum,velec);
 647
 648             fscal            = felec;
 649
 650             fscal            = _mm_and_ps(fscal,cutoff_mask);
 651
 652             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 653
 654             /* Calculate temporary vectorial force */
 655             tx               = _mm_mul_ps(fscal,dx10);
 656             ty               = _mm_mul_ps(fscal,dy10);
 657             tz               = _mm_mul_ps(fscal,dz10);
 658
 659             /* Update vectorial force */
 660             fix1             = _mm_add_ps(fix1,tx);
 661             fiy1             = _mm_add_ps(fiy1,ty);
 662             fiz1             = _mm_add_ps(fiz1,tz);
 663
 664             fjx0             = _mm_add_ps(fjx0,tx);
 665             fjy0             = _mm_add_ps(fjy0,ty);
 666             fjz0             = _mm_add_ps(fjz0,tz);
 667
 668             }
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             if (gmx_mm_any_lt(rsq20,rcutoff2))
 675             {
 676
 677             r20              = _mm_mul_ps(rsq20,rinv20);
 678             r20              = _mm_andnot_ps(dummy_mask,r20);
 679
 680             /* Compute parameters for interactions between i and j atoms */
 681             qq20             = _mm_mul_ps(iq2,jq0);
 682
 683             /* EWALD ELECTROSTATICS */
 684
 685             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 686             ewrt             = _mm_mul_ps(r20,ewtabscale);
 687             ewitab           = _mm_cvttps_epi32(ewrt);
 688             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 689             ewitab           = _mm_slli_epi32(ewitab,2);
 690             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 691             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 692             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 693             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 694             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 695             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 696             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 697             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 698             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 699
 700             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 701
 702             /* Update potential sum for this i atom from the interaction with this j atom. */
 703             velec            = _mm_and_ps(velec,cutoff_mask);
 704             velec            = _mm_andnot_ps(dummy_mask,velec);
 705             velecsum         = _mm_add_ps(velecsum,velec);
 706
 707             fscal            = felec;
 708
 709             fscal            = _mm_and_ps(fscal,cutoff_mask);
 710
 711             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 712
 713             /* Calculate temporary vectorial force */
 714             tx               = _mm_mul_ps(fscal,dx20);
 715             ty               = _mm_mul_ps(fscal,dy20);
 716             tz               = _mm_mul_ps(fscal,dz20);
 717
 718             /* Update vectorial force */
 719             fix2             = _mm_add_ps(fix2,tx);
 720             fiy2             = _mm_add_ps(fiy2,ty);
 721             fiz2             = _mm_add_ps(fiz2,tz);
 722
 723             fjx0             = _mm_add_ps(fjx0,tx);
 724             fjy0             = _mm_add_ps(fjy0,ty);
 725             fjz0             = _mm_add_ps(fjz0,tz);
 726
 727             }
 728
 729             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 730             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 731             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 732             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 733
 734             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 735
 736             /* Inner loop uses 177 flops */
 737         }
 738
 739         /* End of innermost loop */
 740
 741         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 742                                               f+i_coord_offset,fshift+i_shift_offset);
 743
 744         ggid                        = gid[iidx];
 745         /* Update potential energies */
 746         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 747         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 748
 749         /* Increment number of inner iterations */
 750         inneriter                  += j_index_end - j_index_start;
 751
 752         /* Outer loop uses 20 flops */
 753     }
 754
 755     /* Increment number of outer iterations */
 756     outeriter        += nri;
 757
 758     /* Update outer/inner flops */
 759
 760     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*177);
 761 }
 762 /*
 763  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse2_single
 764  * Electrostatics interaction: Ewald
 765  * VdW interaction:            LJEwald
 766  * Geometry:                   Water3-Particle
 767  * Calculate force/pot:        Force
 768  */
 769 void
 770 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse2_single
 771                     (t_nblist                    * gmx_restrict       nlist,
 772                      rvec                        * gmx_restrict          xx,
 773                      rvec                        * gmx_restrict          ff,
 774                      t_forcerec                  * gmx_restrict          fr,
 775                      t_mdatoms                   * gmx_restrict     mdatoms,
 776                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 777                      t_nrnb                      * gmx_restrict        nrnb)
 778 {
 779     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 780      * just 0 for non-waters.
 781      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 782      * jnr indices corresponding to data put in the four positions in the SIMD register.
 783      */
 784     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 785     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 786     int              jnrA,jnrB,jnrC,jnrD;
 787     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 788     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 789     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 790     real             rcutoff_scalar;
 791     real             *shiftvec,*fshift,*x,*f;
 792     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 793     real             scratch[4*DIM];
 794     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 795     int              vdwioffset0;
 796     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 797     int              vdwioffset1;
 798     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 799     int              vdwioffset2;
 800     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 801     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 802     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 803     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 804     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 805     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 806     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 807     real             *charge;
 808     int              nvdwtype;
 809     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 810     int              *vdwtype;
 811     real             *vdwparam;
 812     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 813     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 814     __m128           c6grid_00;
 815     __m128           c6grid_10;
 816     __m128           c6grid_20;
 817     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 818     real             *vdwgridparam;
 819     __m128           one_half = _mm_set1_ps(0.5);
 820     __m128           minus_one = _mm_set1_ps(-1.0);
 821     __m128i          ewitab;
 822     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 823     real             *ewtab;
 824     __m128           dummy_mask,cutoff_mask;
 825     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 826     __m128           one     = _mm_set1_ps(1.0);
 827     __m128           two     = _mm_set1_ps(2.0);
 828     x                = xx[0];
 829     f                = ff[0];
 830
 831     nri              = nlist->nri;
 832     iinr             = nlist->iinr;
 833     jindex           = nlist->jindex;
 834     jjnr             = nlist->jjnr;
 835     shiftidx         = nlist->shift;
 836     gid              = nlist->gid;
 837     shiftvec         = fr->shift_vec[0];
 838     fshift           = fr->fshift[0];
 839     facel            = _mm_set1_ps(fr->epsfac);
 840     charge           = mdatoms->chargeA;
 841     nvdwtype         = fr->ntype;
 842     vdwparam         = fr->nbfp;
 843     vdwtype          = mdatoms->typeA;
 844     vdwgridparam     = fr->ljpme_c6grid;
 845     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 846     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 847     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 848
 849     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 850     ewtab            = fr->ic->tabq_coul_F;
 851     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 852     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 853
 854     /* Setup water-specific parameters */
 855     inr              = nlist->iinr[0];
 856     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 857     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 858     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 859     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 860
 861     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 862     rcutoff_scalar   = fr->rcoulomb;
 863     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 864     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 865
 866     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 867     rvdw             = _mm_set1_ps(fr->rvdw);
 868
 869     /* Avoid stupid compiler warnings */
 870     jnrA = jnrB = jnrC = jnrD = 0;
 871     j_coord_offsetA = 0;
 872     j_coord_offsetB = 0;
 873     j_coord_offsetC = 0;
 874     j_coord_offsetD = 0;
 875
 876     outeriter        = 0;
 877     inneriter        = 0;
 878
 879     for(iidx=0;iidx<4*DIM;iidx++)
 880     {
 881         scratch[iidx] = 0.0;
 882     }
 883
 884     /* Start outer loop over neighborlists */
 885     for(iidx=0; iidx<nri; iidx++)
 886     {
 887         /* Load shift vector for this list */
 888         i_shift_offset   = DIM*shiftidx[iidx];
 889
 890         /* Load limits for loop over neighbors */
 891         j_index_start    = jindex[iidx];
 892         j_index_end      = jindex[iidx+1];
 893
 894         /* Get outer coordinate index */
 895         inr              = iinr[iidx];
 896         i_coord_offset   = DIM*inr;
 897
 898         /* Load i particle coords and add shift vector */
 899         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 900                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 901
 902         fix0             = _mm_setzero_ps();
 903         fiy0             = _mm_setzero_ps();
 904         fiz0             = _mm_setzero_ps();
 905         fix1             = _mm_setzero_ps();
 906         fiy1             = _mm_setzero_ps();
 907         fiz1             = _mm_setzero_ps();
 908         fix2             = _mm_setzero_ps();
 909         fiy2             = _mm_setzero_ps();
 910         fiz2             = _mm_setzero_ps();
 911
 912         /* Start inner kernel loop */
 913         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 914         {
 915
 916             /* Get j neighbor index, and coordinate index */
 917             jnrA             = jjnr[jidx];
 918             jnrB             = jjnr[jidx+1];
 919             jnrC             = jjnr[jidx+2];
 920             jnrD             = jjnr[jidx+3];
 921             j_coord_offsetA  = DIM*jnrA;
 922             j_coord_offsetB  = DIM*jnrB;
 923             j_coord_offsetC  = DIM*jnrC;
 924             j_coord_offsetD  = DIM*jnrD;
 925
 926             /* load j atom coordinates */
 927             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 928                                               x+j_coord_offsetC,x+j_coord_offsetD,
 929                                               &jx0,&jy0,&jz0);
 930
 931             /* Calculate displacement vector */
 932             dx00             = _mm_sub_ps(ix0,jx0);
 933             dy00             = _mm_sub_ps(iy0,jy0);
 934             dz00             = _mm_sub_ps(iz0,jz0);
 935             dx10             = _mm_sub_ps(ix1,jx0);
 936             dy10             = _mm_sub_ps(iy1,jy0);
 937             dz10             = _mm_sub_ps(iz1,jz0);
 938             dx20             = _mm_sub_ps(ix2,jx0);
 939             dy20             = _mm_sub_ps(iy2,jy0);
 940             dz20             = _mm_sub_ps(iz2,jz0);
 941
 942             /* Calculate squared distance and things based on it */
 943             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 944             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 945             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 946
 947             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 948             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 949             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 950
 951             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 952             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 953             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 954
 955             /* Load parameters for j particles */
 956             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 957                                                               charge+jnrC+0,charge+jnrD+0);
 958             vdwjidx0A        = 2*vdwtype[jnrA+0];
 959             vdwjidx0B        = 2*vdwtype[jnrB+0];
 960             vdwjidx0C        = 2*vdwtype[jnrC+0];
 961             vdwjidx0D        = 2*vdwtype[jnrD+0];
 962
 963             fjx0             = _mm_setzero_ps();
 964             fjy0             = _mm_setzero_ps();
 965             fjz0             = _mm_setzero_ps();
 966
 967             /**************************
 968              * CALCULATE INTERACTIONS *
 969              **************************/
 970
 971             if (gmx_mm_any_lt(rsq00,rcutoff2))
 972             {
 973
 974             r00              = _mm_mul_ps(rsq00,rinv00);
 975
 976             /* Compute parameters for interactions between i and j atoms */
 977             qq00             = _mm_mul_ps(iq0,jq0);
 978             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 979                                          vdwparam+vdwioffset0+vdwjidx0B,
 980                                          vdwparam+vdwioffset0+vdwjidx0C,
 981                                          vdwparam+vdwioffset0+vdwjidx0D,
 982                                          &c6_00,&c12_00);
 983             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 984                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 985                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 986                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 987
 988             /* EWALD ELECTROSTATICS */
 989
 990             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 991             ewrt             = _mm_mul_ps(r00,ewtabscale);
 992             ewitab           = _mm_cvttps_epi32(ewrt);
 993             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 994             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 995                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 996                                          &ewtabF,&ewtabFn);
 997             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 998             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 999
1000             /* Analytical LJ-PME */
1001             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1002             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1003             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1004             exponent         = gmx_simd_exp_r(ewcljrsq);
1005             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1006             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1007             /* f6A = 6 * C6grid * (1 - poly) */
1008             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1009             /* f6B = C6grid * exponent * beta^6 */
1010             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1011             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1012             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1013
1014             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1015
1016             fscal            = _mm_add_ps(felec,fvdw);
1017
1018             fscal            = _mm_and_ps(fscal,cutoff_mask);
1019
1020             /* Calculate temporary vectorial force */
1021             tx               = _mm_mul_ps(fscal,dx00);
1022             ty               = _mm_mul_ps(fscal,dy00);
1023             tz               = _mm_mul_ps(fscal,dz00);
1024
1025             /* Update vectorial force */
1026             fix0             = _mm_add_ps(fix0,tx);
1027             fiy0             = _mm_add_ps(fiy0,ty);
1028             fiz0             = _mm_add_ps(fiz0,tz);
1029
1030             fjx0             = _mm_add_ps(fjx0,tx);
1031             fjy0             = _mm_add_ps(fjy0,ty);
1032             fjz0             = _mm_add_ps(fjz0,tz);
1033
1034             }
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             if (gmx_mm_any_lt(rsq10,rcutoff2))
1041             {
1042
1043             r10              = _mm_mul_ps(rsq10,rinv10);
1044
1045             /* Compute parameters for interactions between i and j atoms */
1046             qq10             = _mm_mul_ps(iq1,jq0);
1047
1048             /* EWALD ELECTROSTATICS */
1049
1050             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1051             ewrt             = _mm_mul_ps(r10,ewtabscale);
1052             ewitab           = _mm_cvttps_epi32(ewrt);
1053             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1054             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1055                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1056                                          &ewtabF,&ewtabFn);
1057             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1058             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1059
1060             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1061
1062             fscal            = felec;
1063
1064             fscal            = _mm_and_ps(fscal,cutoff_mask);
1065
1066             /* Calculate temporary vectorial force */
1067             tx               = _mm_mul_ps(fscal,dx10);
1068             ty               = _mm_mul_ps(fscal,dy10);
1069             tz               = _mm_mul_ps(fscal,dz10);
1070
1071             /* Update vectorial force */
1072             fix1             = _mm_add_ps(fix1,tx);
1073             fiy1             = _mm_add_ps(fiy1,ty);
1074             fiz1             = _mm_add_ps(fiz1,tz);
1075
1076             fjx0             = _mm_add_ps(fjx0,tx);
1077             fjy0             = _mm_add_ps(fjy0,ty);
1078             fjz0             = _mm_add_ps(fjz0,tz);
1079
1080             }
1081
1082             /**************************
1083              * CALCULATE INTERACTIONS *
1084              **************************/
1085
1086             if (gmx_mm_any_lt(rsq20,rcutoff2))
1087             {
1088
1089             r20              = _mm_mul_ps(rsq20,rinv20);
1090
1091             /* Compute parameters for interactions between i and j atoms */
1092             qq20             = _mm_mul_ps(iq2,jq0);
1093
1094             /* EWALD ELECTROSTATICS */
1095
1096             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1097             ewrt             = _mm_mul_ps(r20,ewtabscale);
1098             ewitab           = _mm_cvttps_epi32(ewrt);
1099             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1100             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1101                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1102                                          &ewtabF,&ewtabFn);
1103             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1104             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1105
1106             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1107
1108             fscal            = felec;
1109
1110             fscal            = _mm_and_ps(fscal,cutoff_mask);
1111
1112             /* Calculate temporary vectorial force */
1113             tx               = _mm_mul_ps(fscal,dx20);
1114             ty               = _mm_mul_ps(fscal,dy20);
1115             tz               = _mm_mul_ps(fscal,dz20);
1116
1117             /* Update vectorial force */
1118             fix2             = _mm_add_ps(fix2,tx);
1119             fiy2             = _mm_add_ps(fiy2,ty);
1120             fiz2             = _mm_add_ps(fiz2,tz);
1121
1122             fjx0             = _mm_add_ps(fjx0,tx);
1123             fjy0             = _mm_add_ps(fjy0,ty);
1124             fjz0             = _mm_add_ps(fjz0,tz);
1125
1126             }
1127
1128             fjptrA             = f+j_coord_offsetA;
1129             fjptrB             = f+j_coord_offsetB;
1130             fjptrC             = f+j_coord_offsetC;
1131             fjptrD             = f+j_coord_offsetD;
1132
1133             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1134
1135             /* Inner loop uses 140 flops */
1136         }
1137
1138         if(jidx<j_index_end)
1139         {
1140
1141             /* Get j neighbor index, and coordinate index */
1142             jnrlistA         = jjnr[jidx];
1143             jnrlistB         = jjnr[jidx+1];
1144             jnrlistC         = jjnr[jidx+2];
1145             jnrlistD         = jjnr[jidx+3];
1146             /* Sign of each element will be negative for non-real atoms.
1147              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1148              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1149              */
1150             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1151             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1152             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1153             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1154             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1155             j_coord_offsetA  = DIM*jnrA;
1156             j_coord_offsetB  = DIM*jnrB;
1157             j_coord_offsetC  = DIM*jnrC;
1158             j_coord_offsetD  = DIM*jnrD;
1159
1160             /* load j atom coordinates */
1161             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1162                                               x+j_coord_offsetC,x+j_coord_offsetD,
1163                                               &jx0,&jy0,&jz0);
1164
1165             /* Calculate displacement vector */
1166             dx00             = _mm_sub_ps(ix0,jx0);
1167             dy00             = _mm_sub_ps(iy0,jy0);
1168             dz00             = _mm_sub_ps(iz0,jz0);
1169             dx10             = _mm_sub_ps(ix1,jx0);
1170             dy10             = _mm_sub_ps(iy1,jy0);
1171             dz10             = _mm_sub_ps(iz1,jz0);
1172             dx20             = _mm_sub_ps(ix2,jx0);
1173             dy20             = _mm_sub_ps(iy2,jy0);
1174             dz20             = _mm_sub_ps(iz2,jz0);
1175
1176             /* Calculate squared distance and things based on it */
1177             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1178             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1179             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1180
1181             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1182             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1183             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1184
1185             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1186             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1187             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1188
1189             /* Load parameters for j particles */
1190             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1191                                                               charge+jnrC+0,charge+jnrD+0);
1192             vdwjidx0A        = 2*vdwtype[jnrA+0];
1193             vdwjidx0B        = 2*vdwtype[jnrB+0];
1194             vdwjidx0C        = 2*vdwtype[jnrC+0];
1195             vdwjidx0D        = 2*vdwtype[jnrD+0];
1196
1197             fjx0             = _mm_setzero_ps();
1198             fjy0             = _mm_setzero_ps();
1199             fjz0             = _mm_setzero_ps();
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             if (gmx_mm_any_lt(rsq00,rcutoff2))
1206             {
1207
1208             r00              = _mm_mul_ps(rsq00,rinv00);
1209             r00              = _mm_andnot_ps(dummy_mask,r00);
1210
1211             /* Compute parameters for interactions between i and j atoms */
1212             qq00             = _mm_mul_ps(iq0,jq0);
1213             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1214                                          vdwparam+vdwioffset0+vdwjidx0B,
1215                                          vdwparam+vdwioffset0+vdwjidx0C,
1216                                          vdwparam+vdwioffset0+vdwjidx0D,
1217                                          &c6_00,&c12_00);
1218             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1219                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1220                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1221                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1222
1223             /* EWALD ELECTROSTATICS */
1224
1225             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1226             ewrt             = _mm_mul_ps(r00,ewtabscale);
1227             ewitab           = _mm_cvttps_epi32(ewrt);
1228             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1229             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1230                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1231                                          &ewtabF,&ewtabFn);
1232             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1233             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1234
1235             /* Analytical LJ-PME */
1236             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1237             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1238             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1239             exponent         = gmx_simd_exp_r(ewcljrsq);
1240             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1241             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1242             /* f6A = 6 * C6grid * (1 - poly) */
1243             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1244             /* f6B = C6grid * exponent * beta^6 */
1245             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1246             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1247             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1248
1249             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1250
1251             fscal            = _mm_add_ps(felec,fvdw);
1252
1253             fscal            = _mm_and_ps(fscal,cutoff_mask);
1254
1255             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1256
1257             /* Calculate temporary vectorial force */
1258             tx               = _mm_mul_ps(fscal,dx00);
1259             ty               = _mm_mul_ps(fscal,dy00);
1260             tz               = _mm_mul_ps(fscal,dz00);
1261
1262             /* Update vectorial force */
1263             fix0             = _mm_add_ps(fix0,tx);
1264             fiy0             = _mm_add_ps(fiy0,ty);
1265             fiz0             = _mm_add_ps(fiz0,tz);
1266
1267             fjx0             = _mm_add_ps(fjx0,tx);
1268             fjy0             = _mm_add_ps(fjy0,ty);
1269             fjz0             = _mm_add_ps(fjz0,tz);
1270
1271             }
1272
1273             /**************************
1274              * CALCULATE INTERACTIONS *
1275              **************************/
1276
1277             if (gmx_mm_any_lt(rsq10,rcutoff2))
1278             {
1279
1280             r10              = _mm_mul_ps(rsq10,rinv10);
1281             r10              = _mm_andnot_ps(dummy_mask,r10);
1282
1283             /* Compute parameters for interactions between i and j atoms */
1284             qq10             = _mm_mul_ps(iq1,jq0);
1285
1286             /* EWALD ELECTROSTATICS */
1287
1288             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1289             ewrt             = _mm_mul_ps(r10,ewtabscale);
1290             ewitab           = _mm_cvttps_epi32(ewrt);
1291             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1292             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1293                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1294                                          &ewtabF,&ewtabFn);
1295             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1296             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1297
1298             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1299
1300             fscal            = felec;
1301
1302             fscal            = _mm_and_ps(fscal,cutoff_mask);
1303
1304             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1305
1306             /* Calculate temporary vectorial force */
1307             tx               = _mm_mul_ps(fscal,dx10);
1308             ty               = _mm_mul_ps(fscal,dy10);
1309             tz               = _mm_mul_ps(fscal,dz10);
1310
1311             /* Update vectorial force */
1312             fix1             = _mm_add_ps(fix1,tx);
1313             fiy1             = _mm_add_ps(fiy1,ty);
1314             fiz1             = _mm_add_ps(fiz1,tz);
1315
1316             fjx0             = _mm_add_ps(fjx0,tx);
1317             fjy0             = _mm_add_ps(fjy0,ty);
1318             fjz0             = _mm_add_ps(fjz0,tz);
1319
1320             }
1321
1322             /**************************
1323              * CALCULATE INTERACTIONS *
1324              **************************/
1325
1326             if (gmx_mm_any_lt(rsq20,rcutoff2))
1327             {
1328
1329             r20              = _mm_mul_ps(rsq20,rinv20);
1330             r20              = _mm_andnot_ps(dummy_mask,r20);
1331
1332             /* Compute parameters for interactions between i and j atoms */
1333             qq20             = _mm_mul_ps(iq2,jq0);
1334
1335             /* EWALD ELECTROSTATICS */
1336
1337             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1338             ewrt             = _mm_mul_ps(r20,ewtabscale);
1339             ewitab           = _mm_cvttps_epi32(ewrt);
1340             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1341             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1342                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1343                                          &ewtabF,&ewtabFn);
1344             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1345             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1346
1347             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1348
1349             fscal            = felec;
1350
1351             fscal            = _mm_and_ps(fscal,cutoff_mask);
1352
1353             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1354
1355             /* Calculate temporary vectorial force */
1356             tx               = _mm_mul_ps(fscal,dx20);
1357             ty               = _mm_mul_ps(fscal,dy20);
1358             tz               = _mm_mul_ps(fscal,dz20);
1359
1360             /* Update vectorial force */
1361             fix2             = _mm_add_ps(fix2,tx);
1362             fiy2             = _mm_add_ps(fiy2,ty);
1363             fiz2             = _mm_add_ps(fiz2,tz);
1364
1365             fjx0             = _mm_add_ps(fjx0,tx);
1366             fjy0             = _mm_add_ps(fjy0,ty);
1367             fjz0             = _mm_add_ps(fjz0,tz);
1368
1369             }
1370
1371             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1372             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1373             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1374             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1375
1376             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1377
1378             /* Inner loop uses 143 flops */
1379         }
1380
1381         /* End of innermost loop */
1382
1383         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1384                                               f+i_coord_offset,fshift+i_shift_offset);
1385
1386         /* Increment number of inner iterations */
1387         inneriter                  += j_index_end - j_index_start;
1388
1389         /* Outer loop uses 18 flops */
1390     }
1391
1392     /* Increment number of outer iterations */
1393     outeriter        += nri;
1394
1395     /* Update outer/inner flops */
1396
1397     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*143);
1398 }