src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_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_VdwLJEwSh_GeomW3P1_VF_sse4_1_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128           c6grid_00;
 102     __m128           c6grid_10;
 103     __m128           c6grid_20;
 104     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 105     real             *vdwgridparam;
 106     __m128           one_half  = _mm_set1_ps(0.5);
 107     __m128           minus_one = _mm_set1_ps(-1.0);
 108     __m128i          ewitab;
 109     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     real             *ewtab;
 111     __m128           dummy_mask,cutoff_mask;
 112     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 113     __m128           one     = _mm_set1_ps(1.0);
 114     __m128           two     = _mm_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_ps(fr->ic->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131     vdwgridparam     = fr->ljpme_c6grid;
 132     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 133     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 134     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 135
 136     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 144     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 145     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 146     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 149     rcutoff_scalar   = fr->ic->rcoulomb;
 150     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 151     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 152
 153     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 154     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = jnrC = jnrD = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160     j_coord_offsetC = 0;
 161     j_coord_offsetD = 0;
 162
 163     outeriter        = 0;
 164     inneriter        = 0;
 165
 166     for(iidx=0;iidx<4*DIM;iidx++)
 167     {
 168         scratch[iidx] = 0.0;
 169     }
 170
 171     /* Start outer loop over neighborlists */
 172     for(iidx=0; iidx<nri; iidx++)
 173     {
 174         /* Load shift vector for this list */
 175         i_shift_offset   = DIM*shiftidx[iidx];
 176
 177         /* Load limits for loop over neighbors */
 178         j_index_start    = jindex[iidx];
 179         j_index_end      = jindex[iidx+1];
 180
 181         /* Get outer coordinate index */
 182         inr              = iinr[iidx];
 183         i_coord_offset   = DIM*inr;
 184
 185         /* Load i particle coords and add shift vector */
 186         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 187                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 188
 189         fix0             = _mm_setzero_ps();
 190         fiy0             = _mm_setzero_ps();
 191         fiz0             = _mm_setzero_ps();
 192         fix1             = _mm_setzero_ps();
 193         fiy1             = _mm_setzero_ps();
 194         fiz1             = _mm_setzero_ps();
 195         fix2             = _mm_setzero_ps();
 196         fiy2             = _mm_setzero_ps();
 197         fiz2             = _mm_setzero_ps();
 198
 199         /* Reset potential sums */
 200         velecsum         = _mm_setzero_ps();
 201         vvdwsum          = _mm_setzero_ps();
 202
 203         /* Start inner kernel loop */
 204         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 205         {
 206
 207             /* Get j neighbor index, and coordinate index */
 208             jnrA             = jjnr[jidx];
 209             jnrB             = jjnr[jidx+1];
 210             jnrC             = jjnr[jidx+2];
 211             jnrD             = jjnr[jidx+3];
 212             j_coord_offsetA  = DIM*jnrA;
 213             j_coord_offsetB  = DIM*jnrB;
 214             j_coord_offsetC  = DIM*jnrC;
 215             j_coord_offsetD  = DIM*jnrD;
 216
 217             /* load j atom coordinates */
 218             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 219                                               x+j_coord_offsetC,x+j_coord_offsetD,
 220                                               &jx0,&jy0,&jz0);
 221
 222             /* Calculate displacement vector */
 223             dx00             = _mm_sub_ps(ix0,jx0);
 224             dy00             = _mm_sub_ps(iy0,jy0);
 225             dz00             = _mm_sub_ps(iz0,jz0);
 226             dx10             = _mm_sub_ps(ix1,jx0);
 227             dy10             = _mm_sub_ps(iy1,jy0);
 228             dz10             = _mm_sub_ps(iz1,jz0);
 229             dx20             = _mm_sub_ps(ix2,jx0);
 230             dy20             = _mm_sub_ps(iy2,jy0);
 231             dz20             = _mm_sub_ps(iz2,jz0);
 232
 233             /* Calculate squared distance and things based on it */
 234             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 235             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 236             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 237
 238             rinv00           = sse41_invsqrt_f(rsq00);
 239             rinv10           = sse41_invsqrt_f(rsq10);
 240             rinv20           = sse41_invsqrt_f(rsq20);
 241
 242             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 243             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 244             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 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             r00              = _mm_mul_ps(rsq00,rinv00);
 266
 267             /* Compute parameters for interactions between i and j atoms */
 268             qq00             = _mm_mul_ps(iq0,jq0);
 269             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 270                                          vdwparam+vdwioffset0+vdwjidx0B,
 271                                          vdwparam+vdwioffset0+vdwjidx0C,
 272                                          vdwparam+vdwioffset0+vdwjidx0D,
 273                                          &c6_00,&c12_00);
 274
 275             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 276                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 277                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 278                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 279
 280             /* EWALD ELECTROSTATICS */
 281
 282             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 283             ewrt             = _mm_mul_ps(r00,ewtabscale);
 284             ewitab           = _mm_cvttps_epi32(ewrt);
 285             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 286             ewitab           = _mm_slli_epi32(ewitab,2);
 287             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 288             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 289             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 290             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 291             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 292             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 293             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 294             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 295             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 296
 297             /* Analytical LJ-PME */
 298             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 299             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 300             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 301             exponent         = sse41_exp_f(ewcljrsq);
 302             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 303             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 304             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 305             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 306             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 307             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),
 308                                _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));
 309             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 310             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);
 311
 312             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velec            = _mm_and_ps(velec,cutoff_mask);
 316             velecsum         = _mm_add_ps(velecsum,velec);
 317             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 318             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 319
 320             fscal            = _mm_add_ps(felec,fvdw);
 321
 322             fscal            = _mm_and_ps(fscal,cutoff_mask);
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm_mul_ps(fscal,dx00);
 326             ty               = _mm_mul_ps(fscal,dy00);
 327             tz               = _mm_mul_ps(fscal,dz00);
 328
 329             /* Update vectorial force */
 330             fix0             = _mm_add_ps(fix0,tx);
 331             fiy0             = _mm_add_ps(fiy0,ty);
 332             fiz0             = _mm_add_ps(fiz0,tz);
 333
 334             fjx0             = _mm_add_ps(fjx0,tx);
 335             fjy0             = _mm_add_ps(fjy0,ty);
 336             fjz0             = _mm_add_ps(fjz0,tz);
 337
 338             }
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             if (gmx_mm_any_lt(rsq10,rcutoff2))
 345             {
 346
 347             r10              = _mm_mul_ps(rsq10,rinv10);
 348
 349             /* Compute parameters for interactions between i and j atoms */
 350             qq10             = _mm_mul_ps(iq1,jq0);
 351
 352             /* EWALD ELECTROSTATICS */
 353
 354             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 355             ewrt             = _mm_mul_ps(r10,ewtabscale);
 356             ewitab           = _mm_cvttps_epi32(ewrt);
 357             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 358             ewitab           = _mm_slli_epi32(ewitab,2);
 359             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 360             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 361             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 362             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 363             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 364             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 365             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 366             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 367             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 368
 369             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velec            = _mm_and_ps(velec,cutoff_mask);
 373             velecsum         = _mm_add_ps(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377             fscal            = _mm_and_ps(fscal,cutoff_mask);
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm_mul_ps(fscal,dx10);
 381             ty               = _mm_mul_ps(fscal,dy10);
 382             tz               = _mm_mul_ps(fscal,dz10);
 383
 384             /* Update vectorial force */
 385             fix1             = _mm_add_ps(fix1,tx);
 386             fiy1             = _mm_add_ps(fiy1,ty);
 387             fiz1             = _mm_add_ps(fiz1,tz);
 388
 389             fjx0             = _mm_add_ps(fjx0,tx);
 390             fjy0             = _mm_add_ps(fjy0,ty);
 391             fjz0             = _mm_add_ps(fjz0,tz);
 392
 393             }
 394
 395             /**************************
 396              * CALCULATE INTERACTIONS *
 397              **************************/
 398
 399             if (gmx_mm_any_lt(rsq20,rcutoff2))
 400             {
 401
 402             r20              = _mm_mul_ps(rsq20,rinv20);
 403
 404             /* Compute parameters for interactions between i and j atoms */
 405             qq20             = _mm_mul_ps(iq2,jq0);
 406
 407             /* EWALD ELECTROSTATICS */
 408
 409             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 410             ewrt             = _mm_mul_ps(r20,ewtabscale);
 411             ewitab           = _mm_cvttps_epi32(ewrt);
 412             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 413             ewitab           = _mm_slli_epi32(ewitab,2);
 414             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 415             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 416             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 417             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 418             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 419             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 420             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 421             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 422             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 423
 424             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 425
 426             /* Update potential sum for this i atom from the interaction with this j atom. */
 427             velec            = _mm_and_ps(velec,cutoff_mask);
 428             velecsum         = _mm_add_ps(velecsum,velec);
 429
 430             fscal            = felec;
 431
 432             fscal            = _mm_and_ps(fscal,cutoff_mask);
 433
 434             /* Calculate temporary vectorial force */
 435             tx               = _mm_mul_ps(fscal,dx20);
 436             ty               = _mm_mul_ps(fscal,dy20);
 437             tz               = _mm_mul_ps(fscal,dz20);
 438
 439             /* Update vectorial force */
 440             fix2             = _mm_add_ps(fix2,tx);
 441             fiy2             = _mm_add_ps(fiy2,ty);
 442             fiz2             = _mm_add_ps(fiz2,tz);
 443
 444             fjx0             = _mm_add_ps(fjx0,tx);
 445             fjy0             = _mm_add_ps(fjy0,ty);
 446             fjz0             = _mm_add_ps(fjz0,tz);
 447
 448             }
 449
 450             fjptrA             = f+j_coord_offsetA;
 451             fjptrB             = f+j_coord_offsetB;
 452             fjptrC             = f+j_coord_offsetC;
 453             fjptrD             = f+j_coord_offsetD;
 454
 455             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 456
 457             /* Inner loop uses 174 flops */
 458         }
 459
 460         if(jidx<j_index_end)
 461         {
 462
 463             /* Get j neighbor index, and coordinate index */
 464             jnrlistA         = jjnr[jidx];
 465             jnrlistB         = jjnr[jidx+1];
 466             jnrlistC         = jjnr[jidx+2];
 467             jnrlistD         = jjnr[jidx+3];
 468             /* Sign of each element will be negative for non-real atoms.
 469              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 470              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 471              */
 472             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 473             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 474             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 475             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 476             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 477             j_coord_offsetA  = DIM*jnrA;
 478             j_coord_offsetB  = DIM*jnrB;
 479             j_coord_offsetC  = DIM*jnrC;
 480             j_coord_offsetD  = DIM*jnrD;
 481
 482             /* load j atom coordinates */
 483             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 484                                               x+j_coord_offsetC,x+j_coord_offsetD,
 485                                               &jx0,&jy0,&jz0);
 486
 487             /* Calculate displacement vector */
 488             dx00             = _mm_sub_ps(ix0,jx0);
 489             dy00             = _mm_sub_ps(iy0,jy0);
 490             dz00             = _mm_sub_ps(iz0,jz0);
 491             dx10             = _mm_sub_ps(ix1,jx0);
 492             dy10             = _mm_sub_ps(iy1,jy0);
 493             dz10             = _mm_sub_ps(iz1,jz0);
 494             dx20             = _mm_sub_ps(ix2,jx0);
 495             dy20             = _mm_sub_ps(iy2,jy0);
 496             dz20             = _mm_sub_ps(iz2,jz0);
 497
 498             /* Calculate squared distance and things based on it */
 499             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 500             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 501             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 502
 503             rinv00           = sse41_invsqrt_f(rsq00);
 504             rinv10           = sse41_invsqrt_f(rsq10);
 505             rinv20           = sse41_invsqrt_f(rsq20);
 506
 507             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 508             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 509             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 510
 511             /* Load parameters for j particles */
 512             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 513                                                               charge+jnrC+0,charge+jnrD+0);
 514             vdwjidx0A        = 2*vdwtype[jnrA+0];
 515             vdwjidx0B        = 2*vdwtype[jnrB+0];
 516             vdwjidx0C        = 2*vdwtype[jnrC+0];
 517             vdwjidx0D        = 2*vdwtype[jnrD+0];
 518
 519             fjx0             = _mm_setzero_ps();
 520             fjy0             = _mm_setzero_ps();
 521             fjz0             = _mm_setzero_ps();
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             if (gmx_mm_any_lt(rsq00,rcutoff2))
 528             {
 529
 530             r00              = _mm_mul_ps(rsq00,rinv00);
 531             r00              = _mm_andnot_ps(dummy_mask,r00);
 532
 533             /* Compute parameters for interactions between i and j atoms */
 534             qq00             = _mm_mul_ps(iq0,jq0);
 535             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 536                                          vdwparam+vdwioffset0+vdwjidx0B,
 537                                          vdwparam+vdwioffset0+vdwjidx0C,
 538                                          vdwparam+vdwioffset0+vdwjidx0D,
 539                                          &c6_00,&c12_00);
 540
 541             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 542                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 543                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 544                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 545
 546             /* EWALD ELECTROSTATICS */
 547
 548             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 549             ewrt             = _mm_mul_ps(r00,ewtabscale);
 550             ewitab           = _mm_cvttps_epi32(ewrt);
 551             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 552             ewitab           = _mm_slli_epi32(ewitab,2);
 553             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 554             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 555             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 556             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 557             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 558             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 559             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 560             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 561             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 562
 563             /* Analytical LJ-PME */
 564             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 565             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 566             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 567             exponent         = sse41_exp_f(ewcljrsq);
 568             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 569             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 570             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 571             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 572             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 573             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),
 574                                _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));
 575             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 576             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);
 577
 578             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 579
 580             /* Update potential sum for this i atom from the interaction with this j atom. */
 581             velec            = _mm_and_ps(velec,cutoff_mask);
 582             velec            = _mm_andnot_ps(dummy_mask,velec);
 583             velecsum         = _mm_add_ps(velecsum,velec);
 584             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 585             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 586             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 587
 588             fscal            = _mm_add_ps(felec,fvdw);
 589
 590             fscal            = _mm_and_ps(fscal,cutoff_mask);
 591
 592             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm_mul_ps(fscal,dx00);
 596             ty               = _mm_mul_ps(fscal,dy00);
 597             tz               = _mm_mul_ps(fscal,dz00);
 598
 599             /* Update vectorial force */
 600             fix0             = _mm_add_ps(fix0,tx);
 601             fiy0             = _mm_add_ps(fiy0,ty);
 602             fiz0             = _mm_add_ps(fiz0,tz);
 603
 604             fjx0             = _mm_add_ps(fjx0,tx);
 605             fjy0             = _mm_add_ps(fjy0,ty);
 606             fjz0             = _mm_add_ps(fjz0,tz);
 607
 608             }
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             if (gmx_mm_any_lt(rsq10,rcutoff2))
 615             {
 616
 617             r10              = _mm_mul_ps(rsq10,rinv10);
 618             r10              = _mm_andnot_ps(dummy_mask,r10);
 619
 620             /* Compute parameters for interactions between i and j atoms */
 621             qq10             = _mm_mul_ps(iq1,jq0);
 622
 623             /* EWALD ELECTROSTATICS */
 624
 625             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 626             ewrt             = _mm_mul_ps(r10,ewtabscale);
 627             ewitab           = _mm_cvttps_epi32(ewrt);
 628             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 629             ewitab           = _mm_slli_epi32(ewitab,2);
 630             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 631             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 632             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 633             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 634             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 635             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 636             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 637             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 638             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 639
 640             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 641
 642             /* Update potential sum for this i atom from the interaction with this j atom. */
 643             velec            = _mm_and_ps(velec,cutoff_mask);
 644             velec            = _mm_andnot_ps(dummy_mask,velec);
 645             velecsum         = _mm_add_ps(velecsum,velec);
 646
 647             fscal            = felec;
 648
 649             fscal            = _mm_and_ps(fscal,cutoff_mask);
 650
 651             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 652
 653             /* Calculate temporary vectorial force */
 654             tx               = _mm_mul_ps(fscal,dx10);
 655             ty               = _mm_mul_ps(fscal,dy10);
 656             tz               = _mm_mul_ps(fscal,dz10);
 657
 658             /* Update vectorial force */
 659             fix1             = _mm_add_ps(fix1,tx);
 660             fiy1             = _mm_add_ps(fiy1,ty);
 661             fiz1             = _mm_add_ps(fiz1,tz);
 662
 663             fjx0             = _mm_add_ps(fjx0,tx);
 664             fjy0             = _mm_add_ps(fjy0,ty);
 665             fjz0             = _mm_add_ps(fjz0,tz);
 666
 667             }
 668
 669             /**************************
 670              * CALCULATE INTERACTIONS *
 671              **************************/
 672
 673             if (gmx_mm_any_lt(rsq20,rcutoff2))
 674             {
 675
 676             r20              = _mm_mul_ps(rsq20,rinv20);
 677             r20              = _mm_andnot_ps(dummy_mask,r20);
 678
 679             /* Compute parameters for interactions between i and j atoms */
 680             qq20             = _mm_mul_ps(iq2,jq0);
 681
 682             /* EWALD ELECTROSTATICS */
 683
 684             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 685             ewrt             = _mm_mul_ps(r20,ewtabscale);
 686             ewitab           = _mm_cvttps_epi32(ewrt);
 687             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 688             ewitab           = _mm_slli_epi32(ewitab,2);
 689             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 690             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 691             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 692             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 693             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 694             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 695             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 696             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 697             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 698
 699             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 700
 701             /* Update potential sum for this i atom from the interaction with this j atom. */
 702             velec            = _mm_and_ps(velec,cutoff_mask);
 703             velec            = _mm_andnot_ps(dummy_mask,velec);
 704             velecsum         = _mm_add_ps(velecsum,velec);
 705
 706             fscal            = felec;
 707
 708             fscal            = _mm_and_ps(fscal,cutoff_mask);
 709
 710             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 711
 712             /* Calculate temporary vectorial force */
 713             tx               = _mm_mul_ps(fscal,dx20);
 714             ty               = _mm_mul_ps(fscal,dy20);
 715             tz               = _mm_mul_ps(fscal,dz20);
 716
 717             /* Update vectorial force */
 718             fix2             = _mm_add_ps(fix2,tx);
 719             fiy2             = _mm_add_ps(fiy2,ty);
 720             fiz2             = _mm_add_ps(fiz2,tz);
 721
 722             fjx0             = _mm_add_ps(fjx0,tx);
 723             fjy0             = _mm_add_ps(fjy0,ty);
 724             fjz0             = _mm_add_ps(fjz0,tz);
 725
 726             }
 727
 728             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 729             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 730             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 731             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 732
 733             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 734
 735             /* Inner loop uses 177 flops */
 736         }
 737
 738         /* End of innermost loop */
 739
 740         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 741                                               f+i_coord_offset,fshift+i_shift_offset);
 742
 743         ggid                        = gid[iidx];
 744         /* Update potential energies */
 745         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 746         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 747
 748         /* Increment number of inner iterations */
 749         inneriter                  += j_index_end - j_index_start;
 750
 751         /* Outer loop uses 20 flops */
 752     }
 753
 754     /* Increment number of outer iterations */
 755     outeriter        += nri;
 756
 757     /* Update outer/inner flops */
 758
 759     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*177);
 760 }
 761 /*
 762  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse4_1_single
 763  * Electrostatics interaction: Ewald
 764  * VdW interaction:            LJEwald
 765  * Geometry:                   Water3-Particle
 766  * Calculate force/pot:        Force
 767  */
 768 void
 769 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse4_1_single
 770                     (t_nblist                    * gmx_restrict       nlist,
 771                      rvec                        * gmx_restrict          xx,
 772                      rvec                        * gmx_restrict          ff,
 773                      struct t_forcerec           * gmx_restrict          fr,
 774                      t_mdatoms                   * gmx_restrict     mdatoms,
 775                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 776                      t_nrnb                      * gmx_restrict        nrnb)
 777 {
 778     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 779      * just 0 for non-waters.
 780      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 781      * jnr indices corresponding to data put in the four positions in the SIMD register.
 782      */
 783     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 784     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 785     int              jnrA,jnrB,jnrC,jnrD;
 786     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 787     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 788     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 789     real             rcutoff_scalar;
 790     real             *shiftvec,*fshift,*x,*f;
 791     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 792     real             scratch[4*DIM];
 793     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 794     int              vdwioffset0;
 795     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 796     int              vdwioffset1;
 797     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 798     int              vdwioffset2;
 799     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 800     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 801     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 802     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 803     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 804     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 805     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 806     real             *charge;
 807     int              nvdwtype;
 808     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 809     int              *vdwtype;
 810     real             *vdwparam;
 811     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 812     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 813     __m128           c6grid_00;
 814     __m128           c6grid_10;
 815     __m128           c6grid_20;
 816     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 817     real             *vdwgridparam;
 818     __m128           one_half  = _mm_set1_ps(0.5);
 819     __m128           minus_one = _mm_set1_ps(-1.0);
 820     __m128i          ewitab;
 821     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 822     real             *ewtab;
 823     __m128           dummy_mask,cutoff_mask;
 824     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 825     __m128           one     = _mm_set1_ps(1.0);
 826     __m128           two     = _mm_set1_ps(2.0);
 827     x                = xx[0];
 828     f                = ff[0];
 829
 830     nri              = nlist->nri;
 831     iinr             = nlist->iinr;
 832     jindex           = nlist->jindex;
 833     jjnr             = nlist->jjnr;
 834     shiftidx         = nlist->shift;
 835     gid              = nlist->gid;
 836     shiftvec         = fr->shift_vec[0];
 837     fshift           = fr->fshift[0];
 838     facel            = _mm_set1_ps(fr->ic->epsfac);
 839     charge           = mdatoms->chargeA;
 840     nvdwtype         = fr->ntype;
 841     vdwparam         = fr->nbfp;
 842     vdwtype          = mdatoms->typeA;
 843     vdwgridparam     = fr->ljpme_c6grid;
 844     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 845     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 846     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 847
 848     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 849     ewtab            = fr->ic->tabq_coul_F;
 850     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 851     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 852
 853     /* Setup water-specific parameters */
 854     inr              = nlist->iinr[0];
 855     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 856     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 857     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 858     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 859
 860     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 861     rcutoff_scalar   = fr->ic->rcoulomb;
 862     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 863     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 864
 865     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 866     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 867
 868     /* Avoid stupid compiler warnings */
 869     jnrA = jnrB = jnrC = jnrD = 0;
 870     j_coord_offsetA = 0;
 871     j_coord_offsetB = 0;
 872     j_coord_offsetC = 0;
 873     j_coord_offsetD = 0;
 874
 875     outeriter        = 0;
 876     inneriter        = 0;
 877
 878     for(iidx=0;iidx<4*DIM;iidx++)
 879     {
 880         scratch[iidx] = 0.0;
 881     }
 882
 883     /* Start outer loop over neighborlists */
 884     for(iidx=0; iidx<nri; iidx++)
 885     {
 886         /* Load shift vector for this list */
 887         i_shift_offset   = DIM*shiftidx[iidx];
 888
 889         /* Load limits for loop over neighbors */
 890         j_index_start    = jindex[iidx];
 891         j_index_end      = jindex[iidx+1];
 892
 893         /* Get outer coordinate index */
 894         inr              = iinr[iidx];
 895         i_coord_offset   = DIM*inr;
 896
 897         /* Load i particle coords and add shift vector */
 898         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 899                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 900
 901         fix0             = _mm_setzero_ps();
 902         fiy0             = _mm_setzero_ps();
 903         fiz0             = _mm_setzero_ps();
 904         fix1             = _mm_setzero_ps();
 905         fiy1             = _mm_setzero_ps();
 906         fiz1             = _mm_setzero_ps();
 907         fix2             = _mm_setzero_ps();
 908         fiy2             = _mm_setzero_ps();
 909         fiz2             = _mm_setzero_ps();
 910
 911         /* Start inner kernel loop */
 912         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 913         {
 914
 915             /* Get j neighbor index, and coordinate index */
 916             jnrA             = jjnr[jidx];
 917             jnrB             = jjnr[jidx+1];
 918             jnrC             = jjnr[jidx+2];
 919             jnrD             = jjnr[jidx+3];
 920             j_coord_offsetA  = DIM*jnrA;
 921             j_coord_offsetB  = DIM*jnrB;
 922             j_coord_offsetC  = DIM*jnrC;
 923             j_coord_offsetD  = DIM*jnrD;
 924
 925             /* load j atom coordinates */
 926             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 927                                               x+j_coord_offsetC,x+j_coord_offsetD,
 928                                               &jx0,&jy0,&jz0);
 929
 930             /* Calculate displacement vector */
 931             dx00             = _mm_sub_ps(ix0,jx0);
 932             dy00             = _mm_sub_ps(iy0,jy0);
 933             dz00             = _mm_sub_ps(iz0,jz0);
 934             dx10             = _mm_sub_ps(ix1,jx0);
 935             dy10             = _mm_sub_ps(iy1,jy0);
 936             dz10             = _mm_sub_ps(iz1,jz0);
 937             dx20             = _mm_sub_ps(ix2,jx0);
 938             dy20             = _mm_sub_ps(iy2,jy0);
 939             dz20             = _mm_sub_ps(iz2,jz0);
 940
 941             /* Calculate squared distance and things based on it */
 942             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 943             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 944             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 945
 946             rinv00           = sse41_invsqrt_f(rsq00);
 947             rinv10           = sse41_invsqrt_f(rsq10);
 948             rinv20           = sse41_invsqrt_f(rsq20);
 949
 950             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 951             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 952             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 953
 954             /* Load parameters for j particles */
 955             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 956                                                               charge+jnrC+0,charge+jnrD+0);
 957             vdwjidx0A        = 2*vdwtype[jnrA+0];
 958             vdwjidx0B        = 2*vdwtype[jnrB+0];
 959             vdwjidx0C        = 2*vdwtype[jnrC+0];
 960             vdwjidx0D        = 2*vdwtype[jnrD+0];
 961
 962             fjx0             = _mm_setzero_ps();
 963             fjy0             = _mm_setzero_ps();
 964             fjz0             = _mm_setzero_ps();
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             if (gmx_mm_any_lt(rsq00,rcutoff2))
 971             {
 972
 973             r00              = _mm_mul_ps(rsq00,rinv00);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq00             = _mm_mul_ps(iq0,jq0);
 977             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 978                                          vdwparam+vdwioffset0+vdwjidx0B,
 979                                          vdwparam+vdwioffset0+vdwjidx0C,
 980                                          vdwparam+vdwioffset0+vdwjidx0D,
 981                                          &c6_00,&c12_00);
 982
 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_round_ps(ewrt, _MM_FROUND_FLOOR));
 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         = sse41_exp_f(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_round_ps(ewrt, _MM_FROUND_FLOOR));
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_round_ps(ewrt, _MM_FROUND_FLOOR));
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           = sse41_invsqrt_f(rsq00);
1182             rinv10           = sse41_invsqrt_f(rsq10);
1183             rinv20           = sse41_invsqrt_f(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
1219             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1220                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1221                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1222                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1223
1224             /* EWALD ELECTROSTATICS */
1225
1226             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1227             ewrt             = _mm_mul_ps(r00,ewtabscale);
1228             ewitab           = _mm_cvttps_epi32(ewrt);
1229             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1230             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1231                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1232                                          &ewtabF,&ewtabFn);
1233             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1234             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1235
1236             /* Analytical LJ-PME */
1237             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1238             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1239             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1240             exponent         = sse41_exp_f(ewcljrsq);
1241             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1242             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1243             /* f6A = 6 * C6grid * (1 - poly) */
1244             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1245             /* f6B = C6grid * exponent * beta^6 */
1246             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1247             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1248             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);
1249
1250             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1251
1252             fscal            = _mm_add_ps(felec,fvdw);
1253
1254             fscal            = _mm_and_ps(fscal,cutoff_mask);
1255
1256             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1257
1258             /* Calculate temporary vectorial force */
1259             tx               = _mm_mul_ps(fscal,dx00);
1260             ty               = _mm_mul_ps(fscal,dy00);
1261             tz               = _mm_mul_ps(fscal,dz00);
1262
1263             /* Update vectorial force */
1264             fix0             = _mm_add_ps(fix0,tx);
1265             fiy0             = _mm_add_ps(fiy0,ty);
1266             fiz0             = _mm_add_ps(fiz0,tz);
1267
1268             fjx0             = _mm_add_ps(fjx0,tx);
1269             fjy0             = _mm_add_ps(fjy0,ty);
1270             fjz0             = _mm_add_ps(fjz0,tz);
1271
1272             }
1273
1274             /**************************
1275              * CALCULATE INTERACTIONS *
1276              **************************/
1277
1278             if (gmx_mm_any_lt(rsq10,rcutoff2))
1279             {
1280
1281             r10              = _mm_mul_ps(rsq10,rinv10);
1282             r10              = _mm_andnot_ps(dummy_mask,r10);
1283
1284             /* Compute parameters for interactions between i and j atoms */
1285             qq10             = _mm_mul_ps(iq1,jq0);
1286
1287             /* EWALD ELECTROSTATICS */
1288
1289             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1290             ewrt             = _mm_mul_ps(r10,ewtabscale);
1291             ewitab           = _mm_cvttps_epi32(ewrt);
1292             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1293             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1294                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1295                                          &ewtabF,&ewtabFn);
1296             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1297             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1298
1299             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1300
1301             fscal            = felec;
1302
1303             fscal            = _mm_and_ps(fscal,cutoff_mask);
1304
1305             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1306
1307             /* Calculate temporary vectorial force */
1308             tx               = _mm_mul_ps(fscal,dx10);
1309             ty               = _mm_mul_ps(fscal,dy10);
1310             tz               = _mm_mul_ps(fscal,dz10);
1311
1312             /* Update vectorial force */
1313             fix1             = _mm_add_ps(fix1,tx);
1314             fiy1             = _mm_add_ps(fiy1,ty);
1315             fiz1             = _mm_add_ps(fiz1,tz);
1316
1317             fjx0             = _mm_add_ps(fjx0,tx);
1318             fjy0             = _mm_add_ps(fjy0,ty);
1319             fjz0             = _mm_add_ps(fjz0,tz);
1320
1321             }
1322
1323             /**************************
1324              * CALCULATE INTERACTIONS *
1325              **************************/
1326
1327             if (gmx_mm_any_lt(rsq20,rcutoff2))
1328             {
1329
1330             r20              = _mm_mul_ps(rsq20,rinv20);
1331             r20              = _mm_andnot_ps(dummy_mask,r20);
1332
1333             /* Compute parameters for interactions between i and j atoms */
1334             qq20             = _mm_mul_ps(iq2,jq0);
1335
1336             /* EWALD ELECTROSTATICS */
1337
1338             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1339             ewrt             = _mm_mul_ps(r20,ewtabscale);
1340             ewitab           = _mm_cvttps_epi32(ewrt);
1341             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1342             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1343                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1344                                          &ewtabF,&ewtabFn);
1345             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1346             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1347
1348             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1349
1350             fscal            = felec;
1351
1352             fscal            = _mm_and_ps(fscal,cutoff_mask);
1353
1354             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1355
1356             /* Calculate temporary vectorial force */
1357             tx               = _mm_mul_ps(fscal,dx20);
1358             ty               = _mm_mul_ps(fscal,dy20);
1359             tz               = _mm_mul_ps(fscal,dz20);
1360
1361             /* Update vectorial force */
1362             fix2             = _mm_add_ps(fix2,tx);
1363             fiy2             = _mm_add_ps(fiy2,ty);
1364             fiz2             = _mm_add_ps(fiz2,tz);
1365
1366             fjx0             = _mm_add_ps(fjx0,tx);
1367             fjy0             = _mm_add_ps(fjy0,ty);
1368             fjz0             = _mm_add_ps(fjz0,tz);
1369
1370             }
1371
1372             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1373             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1374             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1375             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1376
1377             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1378
1379             /* Inner loop uses 143 flops */
1380         }
1381
1382         /* End of innermost loop */
1383
1384         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1385                                               f+i_coord_offset,fshift+i_shift_offset);
1386
1387         /* Increment number of inner iterations */
1388         inneriter                  += j_index_end - j_index_start;
1389
1390         /* Outer loop uses 18 flops */
1391     }
1392
1393     /* Increment number of outer iterations */
1394     outeriter        += nri;
1395
1396     /* Update outer/inner flops */
1397
1398     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*143);
1399 }