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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128           c6grid_00;
 103     __m128           c6grid_10;
 104     __m128           c6grid_20;
 105     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 106     real             *vdwgridparam;
 107     __m128           one_half = _mm_set1_ps(0.5);
 108     __m128           minus_one = _mm_set1_ps(-1.0);
 109     __m128i          ewitab;
 110     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     real             *ewtab;
 112     __m128           dummy_mask,cutoff_mask;
 113     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 114     __m128           one     = _mm_set1_ps(1.0);
 115     __m128           two     = _mm_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm_set1_ps(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132     vdwgridparam     = fr->ljpme_c6grid;
 133     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 134     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 135     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 136
 137     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 138     ewtab            = fr->ic->tabq_coul_FDV0;
 139     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 140     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 145     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 146     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 147     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 150     rcutoff_scalar   = fr->rcoulomb;
 151     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 152     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 153
 154     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 155     rvdw             = _mm_set1_ps(fr->rvdw);
 156
 157     /* Avoid stupid compiler warnings */
 158     jnrA = jnrB = jnrC = jnrD = 0;
 159     j_coord_offsetA = 0;
 160     j_coord_offsetB = 0;
 161     j_coord_offsetC = 0;
 162     j_coord_offsetD = 0;
 163
 164     outeriter        = 0;
 165     inneriter        = 0;
 166
 167     for(iidx=0;iidx<4*DIM;iidx++)
 168     {
 169         scratch[iidx] = 0.0;
 170     }
 171
 172     /* Start outer loop over neighborlists */
 173     for(iidx=0; iidx<nri; iidx++)
 174     {
 175         /* Load shift vector for this list */
 176         i_shift_offset   = DIM*shiftidx[iidx];
 177
 178         /* Load limits for loop over neighbors */
 179         j_index_start    = jindex[iidx];
 180         j_index_end      = jindex[iidx+1];
 181
 182         /* Get outer coordinate index */
 183         inr              = iinr[iidx];
 184         i_coord_offset   = DIM*inr;
 185
 186         /* Load i particle coords and add shift vector */
 187         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 188                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 189
 190         fix0             = _mm_setzero_ps();
 191         fiy0             = _mm_setzero_ps();
 192         fiz0             = _mm_setzero_ps();
 193         fix1             = _mm_setzero_ps();
 194         fiy1             = _mm_setzero_ps();
 195         fiz1             = _mm_setzero_ps();
 196         fix2             = _mm_setzero_ps();
 197         fiy2             = _mm_setzero_ps();
 198         fiz2             = _mm_setzero_ps();
 199
 200         /* Reset potential sums */
 201         velecsum         = _mm_setzero_ps();
 202         vvdwsum          = _mm_setzero_ps();
 203
 204         /* Start inner kernel loop */
 205         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 206         {
 207
 208             /* Get j neighbor index, and coordinate index */
 209             jnrA             = jjnr[jidx];
 210             jnrB             = jjnr[jidx+1];
 211             jnrC             = jjnr[jidx+2];
 212             jnrD             = jjnr[jidx+3];
 213             j_coord_offsetA  = DIM*jnrA;
 214             j_coord_offsetB  = DIM*jnrB;
 215             j_coord_offsetC  = DIM*jnrC;
 216             j_coord_offsetD  = DIM*jnrD;
 217
 218             /* load j atom coordinates */
 219             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 220                                               x+j_coord_offsetC,x+j_coord_offsetD,
 221                                               &jx0,&jy0,&jz0);
 222
 223             /* Calculate displacement vector */
 224             dx00             = _mm_sub_ps(ix0,jx0);
 225             dy00             = _mm_sub_ps(iy0,jy0);
 226             dz00             = _mm_sub_ps(iz0,jz0);
 227             dx10             = _mm_sub_ps(ix1,jx0);
 228             dy10             = _mm_sub_ps(iy1,jy0);
 229             dz10             = _mm_sub_ps(iz1,jz0);
 230             dx20             = _mm_sub_ps(ix2,jx0);
 231             dy20             = _mm_sub_ps(iy2,jy0);
 232             dz20             = _mm_sub_ps(iz2,jz0);
 233
 234             /* Calculate squared distance and things based on it */
 235             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 236             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 237             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 238
 239             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 240             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 241             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 242
 243             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 244             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 245             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 246
 247             /* Load parameters for j particles */
 248             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 249                                                               charge+jnrC+0,charge+jnrD+0);
 250             vdwjidx0A        = 2*vdwtype[jnrA+0];
 251             vdwjidx0B        = 2*vdwtype[jnrB+0];
 252             vdwjidx0C        = 2*vdwtype[jnrC+0];
 253             vdwjidx0D        = 2*vdwtype[jnrD+0];
 254
 255             fjx0             = _mm_setzero_ps();
 256             fjy0             = _mm_setzero_ps();
 257             fjz0             = _mm_setzero_ps();
 258
 259             /**************************
 260              * CALCULATE INTERACTIONS *
 261              **************************/
 262
 263             if (gmx_mm_any_lt(rsq00,rcutoff2))
 264             {
 265
 266             r00              = _mm_mul_ps(rsq00,rinv00);
 267
 268             /* Compute parameters for interactions between i and j atoms */
 269             qq00             = _mm_mul_ps(iq0,jq0);
 270             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 271                                          vdwparam+vdwioffset0+vdwjidx0B,
 272                                          vdwparam+vdwioffset0+vdwjidx0C,
 273                                          vdwparam+vdwioffset0+vdwjidx0D,
 274                                          &c6_00,&c12_00);
 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_cvtepi32_ps(ewitab));
 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         = gmx_simd_exp_r(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_cvtepi32_ps(ewitab));
 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_cvtepi32_ps(ewitab));
 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           = gmx_mm_invsqrt_ps(rsq00);
 504             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 505             rinv20           = gmx_mm_invsqrt_ps(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             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 541                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 542                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 543                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 544
 545             /* EWALD ELECTROSTATICS */
 546
 547             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 548             ewrt             = _mm_mul_ps(r00,ewtabscale);
 549             ewitab           = _mm_cvttps_epi32(ewrt);
 550             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 551             ewitab           = _mm_slli_epi32(ewitab,2);
 552             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 553             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 554             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 555             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 556             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 557             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 558             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 559             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 560             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 561
 562             /* Analytical LJ-PME */
 563             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 564             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 565             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 566             exponent         = gmx_simd_exp_r(ewcljrsq);
 567             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 568             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 569             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 570             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 571             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 572             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) ,
 573                                           _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));
 574             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 575             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);
 576
 577             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             velec            = _mm_and_ps(velec,cutoff_mask);
 581             velec            = _mm_andnot_ps(dummy_mask,velec);
 582             velecsum         = _mm_add_ps(velecsum,velec);
 583             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 584             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 585             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 586
 587             fscal            = _mm_add_ps(felec,fvdw);
 588
 589             fscal            = _mm_and_ps(fscal,cutoff_mask);
 590
 591             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm_mul_ps(fscal,dx00);
 595             ty               = _mm_mul_ps(fscal,dy00);
 596             tz               = _mm_mul_ps(fscal,dz00);
 597
 598             /* Update vectorial force */
 599             fix0             = _mm_add_ps(fix0,tx);
 600             fiy0             = _mm_add_ps(fiy0,ty);
 601             fiz0             = _mm_add_ps(fiz0,tz);
 602
 603             fjx0             = _mm_add_ps(fjx0,tx);
 604             fjy0             = _mm_add_ps(fjy0,ty);
 605             fjz0             = _mm_add_ps(fjz0,tz);
 606
 607             }
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             if (gmx_mm_any_lt(rsq10,rcutoff2))
 614             {
 615
 616             r10              = _mm_mul_ps(rsq10,rinv10);
 617             r10              = _mm_andnot_ps(dummy_mask,r10);
 618
 619             /* Compute parameters for interactions between i and j atoms */
 620             qq10             = _mm_mul_ps(iq1,jq0);
 621
 622             /* EWALD ELECTROSTATICS */
 623
 624             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 625             ewrt             = _mm_mul_ps(r10,ewtabscale);
 626             ewitab           = _mm_cvttps_epi32(ewrt);
 627             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 628             ewitab           = _mm_slli_epi32(ewitab,2);
 629             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 630             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 631             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 632             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 633             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 634             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 635             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 636             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 637             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 638
 639             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 640
 641             /* Update potential sum for this i atom from the interaction with this j atom. */
 642             velec            = _mm_and_ps(velec,cutoff_mask);
 643             velec            = _mm_andnot_ps(dummy_mask,velec);
 644             velecsum         = _mm_add_ps(velecsum,velec);
 645
 646             fscal            = felec;
 647
 648             fscal            = _mm_and_ps(fscal,cutoff_mask);
 649
 650             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 651
 652             /* Calculate temporary vectorial force */
 653             tx               = _mm_mul_ps(fscal,dx10);
 654             ty               = _mm_mul_ps(fscal,dy10);
 655             tz               = _mm_mul_ps(fscal,dz10);
 656
 657             /* Update vectorial force */
 658             fix1             = _mm_add_ps(fix1,tx);
 659             fiy1             = _mm_add_ps(fiy1,ty);
 660             fiz1             = _mm_add_ps(fiz1,tz);
 661
 662             fjx0             = _mm_add_ps(fjx0,tx);
 663             fjy0             = _mm_add_ps(fjy0,ty);
 664             fjz0             = _mm_add_ps(fjz0,tz);
 665
 666             }
 667
 668             /**************************
 669              * CALCULATE INTERACTIONS *
 670              **************************/
 671
 672             if (gmx_mm_any_lt(rsq20,rcutoff2))
 673             {
 674
 675             r20              = _mm_mul_ps(rsq20,rinv20);
 676             r20              = _mm_andnot_ps(dummy_mask,r20);
 677
 678             /* Compute parameters for interactions between i and j atoms */
 679             qq20             = _mm_mul_ps(iq2,jq0);
 680
 681             /* EWALD ELECTROSTATICS */
 682
 683             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 684             ewrt             = _mm_mul_ps(r20,ewtabscale);
 685             ewitab           = _mm_cvttps_epi32(ewrt);
 686             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 687             ewitab           = _mm_slli_epi32(ewitab,2);
 688             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 689             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 690             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 691             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 692             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 693             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 694             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 695             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 696             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 697
 698             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 699
 700             /* Update potential sum for this i atom from the interaction with this j atom. */
 701             velec            = _mm_and_ps(velec,cutoff_mask);
 702             velec            = _mm_andnot_ps(dummy_mask,velec);
 703             velecsum         = _mm_add_ps(velecsum,velec);
 704
 705             fscal            = felec;
 706
 707             fscal            = _mm_and_ps(fscal,cutoff_mask);
 708
 709             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 710
 711             /* Calculate temporary vectorial force */
 712             tx               = _mm_mul_ps(fscal,dx20);
 713             ty               = _mm_mul_ps(fscal,dy20);
 714             tz               = _mm_mul_ps(fscal,dz20);
 715
 716             /* Update vectorial force */
 717             fix2             = _mm_add_ps(fix2,tx);
 718             fiy2             = _mm_add_ps(fiy2,ty);
 719             fiz2             = _mm_add_ps(fiz2,tz);
 720
 721             fjx0             = _mm_add_ps(fjx0,tx);
 722             fjy0             = _mm_add_ps(fjy0,ty);
 723             fjz0             = _mm_add_ps(fjz0,tz);
 724
 725             }
 726
 727             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 728             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 729             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 730             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 731
 732             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 733
 734             /* Inner loop uses 177 flops */
 735         }
 736
 737         /* End of innermost loop */
 738
 739         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 740                                               f+i_coord_offset,fshift+i_shift_offset);
 741
 742         ggid                        = gid[iidx];
 743         /* Update potential energies */
 744         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 745         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 746
 747         /* Increment number of inner iterations */
 748         inneriter                  += j_index_end - j_index_start;
 749
 750         /* Outer loop uses 20 flops */
 751     }
 752
 753     /* Increment number of outer iterations */
 754     outeriter        += nri;
 755
 756     /* Update outer/inner flops */
 757
 758     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*177);
 759 }
 760 /*
 761  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse2_single
 762  * Electrostatics interaction: Ewald
 763  * VdW interaction:            LJEwald
 764  * Geometry:                   Water3-Particle
 765  * Calculate force/pot:        Force
 766  */
 767 void
 768 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse2_single
 769                     (t_nblist                    * gmx_restrict       nlist,
 770                      rvec                        * gmx_restrict          xx,
 771                      rvec                        * gmx_restrict          ff,
 772                      t_forcerec                  * gmx_restrict          fr,
 773                      t_mdatoms                   * gmx_restrict     mdatoms,
 774                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 775                      t_nrnb                      * gmx_restrict        nrnb)
 776 {
 777     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 778      * just 0 for non-waters.
 779      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 780      * jnr indices corresponding to data put in the four positions in the SIMD register.
 781      */
 782     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 783     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 784     int              jnrA,jnrB,jnrC,jnrD;
 785     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 786     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 787     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 788     real             rcutoff_scalar;
 789     real             *shiftvec,*fshift,*x,*f;
 790     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 791     real             scratch[4*DIM];
 792     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 793     int              vdwioffset0;
 794     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 795     int              vdwioffset1;
 796     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 797     int              vdwioffset2;
 798     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 799     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 800     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 801     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 802     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 803     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 804     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 805     real             *charge;
 806     int              nvdwtype;
 807     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 808     int              *vdwtype;
 809     real             *vdwparam;
 810     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 811     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 812     __m128           c6grid_00;
 813     __m128           c6grid_10;
 814     __m128           c6grid_20;
 815     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 816     real             *vdwgridparam;
 817     __m128           one_half = _mm_set1_ps(0.5);
 818     __m128           minus_one = _mm_set1_ps(-1.0);
 819     __m128i          ewitab;
 820     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 821     real             *ewtab;
 822     __m128           dummy_mask,cutoff_mask;
 823     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 824     __m128           one     = _mm_set1_ps(1.0);
 825     __m128           two     = _mm_set1_ps(2.0);
 826     x                = xx[0];
 827     f                = ff[0];
 828
 829     nri              = nlist->nri;
 830     iinr             = nlist->iinr;
 831     jindex           = nlist->jindex;
 832     jjnr             = nlist->jjnr;
 833     shiftidx         = nlist->shift;
 834     gid              = nlist->gid;
 835     shiftvec         = fr->shift_vec[0];
 836     fshift           = fr->fshift[0];
 837     facel            = _mm_set1_ps(fr->epsfac);
 838     charge           = mdatoms->chargeA;
 839     nvdwtype         = fr->ntype;
 840     vdwparam         = fr->nbfp;
 841     vdwtype          = mdatoms->typeA;
 842     vdwgridparam     = fr->ljpme_c6grid;
 843     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 844     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 845     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 846
 847     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 848     ewtab            = fr->ic->tabq_coul_F;
 849     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 850     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 851
 852     /* Setup water-specific parameters */
 853     inr              = nlist->iinr[0];
 854     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 855     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 856     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 857     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 858
 859     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 860     rcutoff_scalar   = fr->rcoulomb;
 861     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 862     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 863
 864     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 865     rvdw             = _mm_set1_ps(fr->rvdw);
 866
 867     /* Avoid stupid compiler warnings */
 868     jnrA = jnrB = jnrC = jnrD = 0;
 869     j_coord_offsetA = 0;
 870     j_coord_offsetB = 0;
 871     j_coord_offsetC = 0;
 872     j_coord_offsetD = 0;
 873
 874     outeriter        = 0;
 875     inneriter        = 0;
 876
 877     for(iidx=0;iidx<4*DIM;iidx++)
 878     {
 879         scratch[iidx] = 0.0;
 880     }
 881
 882     /* Start outer loop over neighborlists */
 883     for(iidx=0; iidx<nri; iidx++)
 884     {
 885         /* Load shift vector for this list */
 886         i_shift_offset   = DIM*shiftidx[iidx];
 887
 888         /* Load limits for loop over neighbors */
 889         j_index_start    = jindex[iidx];
 890         j_index_end      = jindex[iidx+1];
 891
 892         /* Get outer coordinate index */
 893         inr              = iinr[iidx];
 894         i_coord_offset   = DIM*inr;
 895
 896         /* Load i particle coords and add shift vector */
 897         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 898                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 899
 900         fix0             = _mm_setzero_ps();
 901         fiy0             = _mm_setzero_ps();
 902         fiz0             = _mm_setzero_ps();
 903         fix1             = _mm_setzero_ps();
 904         fiy1             = _mm_setzero_ps();
 905         fiz1             = _mm_setzero_ps();
 906         fix2             = _mm_setzero_ps();
 907         fiy2             = _mm_setzero_ps();
 908         fiz2             = _mm_setzero_ps();
 909
 910         /* Start inner kernel loop */
 911         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 912         {
 913
 914             /* Get j neighbor index, and coordinate index */
 915             jnrA             = jjnr[jidx];
 916             jnrB             = jjnr[jidx+1];
 917             jnrC             = jjnr[jidx+2];
 918             jnrD             = jjnr[jidx+3];
 919             j_coord_offsetA  = DIM*jnrA;
 920             j_coord_offsetB  = DIM*jnrB;
 921             j_coord_offsetC  = DIM*jnrC;
 922             j_coord_offsetD  = DIM*jnrD;
 923
 924             /* load j atom coordinates */
 925             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 926                                               x+j_coord_offsetC,x+j_coord_offsetD,
 927                                               &jx0,&jy0,&jz0);
 928
 929             /* Calculate displacement vector */
 930             dx00             = _mm_sub_ps(ix0,jx0);
 931             dy00             = _mm_sub_ps(iy0,jy0);
 932             dz00             = _mm_sub_ps(iz0,jz0);
 933             dx10             = _mm_sub_ps(ix1,jx0);
 934             dy10             = _mm_sub_ps(iy1,jy0);
 935             dz10             = _mm_sub_ps(iz1,jz0);
 936             dx20             = _mm_sub_ps(ix2,jx0);
 937             dy20             = _mm_sub_ps(iy2,jy0);
 938             dz20             = _mm_sub_ps(iz2,jz0);
 939
 940             /* Calculate squared distance and things based on it */
 941             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 942             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 943             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 944
 945             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 946             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 947             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 948
 949             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 950             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 951             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 952
 953             /* Load parameters for j particles */
 954             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 955                                                               charge+jnrC+0,charge+jnrD+0);
 956             vdwjidx0A        = 2*vdwtype[jnrA+0];
 957             vdwjidx0B        = 2*vdwtype[jnrB+0];
 958             vdwjidx0C        = 2*vdwtype[jnrC+0];
 959             vdwjidx0D        = 2*vdwtype[jnrD+0];
 960
 961             fjx0             = _mm_setzero_ps();
 962             fjy0             = _mm_setzero_ps();
 963             fjz0             = _mm_setzero_ps();
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             if (gmx_mm_any_lt(rsq00,rcutoff2))
 970             {
 971
 972             r00              = _mm_mul_ps(rsq00,rinv00);
 973
 974             /* Compute parameters for interactions between i and j atoms */
 975             qq00             = _mm_mul_ps(iq0,jq0);
 976             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 977                                          vdwparam+vdwioffset0+vdwjidx0B,
 978                                          vdwparam+vdwioffset0+vdwjidx0C,
 979                                          vdwparam+vdwioffset0+vdwjidx0D,
 980                                          &c6_00,&c12_00);
 981             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 982                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 983                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 984                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 985
 986             /* EWALD ELECTROSTATICS */
 987
 988             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 989             ewrt             = _mm_mul_ps(r00,ewtabscale);
 990             ewitab           = _mm_cvttps_epi32(ewrt);
 991             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 992             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 993                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 994                                          &ewtabF,&ewtabFn);
 995             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 996             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 997
 998             /* Analytical LJ-PME */
 999             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1000             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1001             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1002             exponent         = gmx_simd_exp_r(ewcljrsq);
1003             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1004             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1005             /* f6A = 6 * C6grid * (1 - poly) */
1006             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1007             /* f6B = C6grid * exponent * beta^6 */
1008             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1009             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1010             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);
1011
1012             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1013
1014             fscal            = _mm_add_ps(felec,fvdw);
1015
1016             fscal            = _mm_and_ps(fscal,cutoff_mask);
1017
1018             /* Calculate temporary vectorial force */
1019             tx               = _mm_mul_ps(fscal,dx00);
1020             ty               = _mm_mul_ps(fscal,dy00);
1021             tz               = _mm_mul_ps(fscal,dz00);
1022
1023             /* Update vectorial force */
1024             fix0             = _mm_add_ps(fix0,tx);
1025             fiy0             = _mm_add_ps(fiy0,ty);
1026             fiz0             = _mm_add_ps(fiz0,tz);
1027
1028             fjx0             = _mm_add_ps(fjx0,tx);
1029             fjy0             = _mm_add_ps(fjy0,ty);
1030             fjz0             = _mm_add_ps(fjz0,tz);
1031
1032             }
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             if (gmx_mm_any_lt(rsq10,rcutoff2))
1039             {
1040
1041             r10              = _mm_mul_ps(rsq10,rinv10);
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             qq10             = _mm_mul_ps(iq1,jq0);
1045
1046             /* EWALD ELECTROSTATICS */
1047
1048             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1049             ewrt             = _mm_mul_ps(r10,ewtabscale);
1050             ewitab           = _mm_cvttps_epi32(ewrt);
1051             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1052             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1053                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1054                                          &ewtabF,&ewtabFn);
1055             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1056             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1057
1058             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1059
1060             fscal            = felec;
1061
1062             fscal            = _mm_and_ps(fscal,cutoff_mask);
1063
1064             /* Calculate temporary vectorial force */
1065             tx               = _mm_mul_ps(fscal,dx10);
1066             ty               = _mm_mul_ps(fscal,dy10);
1067             tz               = _mm_mul_ps(fscal,dz10);
1068
1069             /* Update vectorial force */
1070             fix1             = _mm_add_ps(fix1,tx);
1071             fiy1             = _mm_add_ps(fiy1,ty);
1072             fiz1             = _mm_add_ps(fiz1,tz);
1073
1074             fjx0             = _mm_add_ps(fjx0,tx);
1075             fjy0             = _mm_add_ps(fjy0,ty);
1076             fjz0             = _mm_add_ps(fjz0,tz);
1077
1078             }
1079
1080             /**************************
1081              * CALCULATE INTERACTIONS *
1082              **************************/
1083
1084             if (gmx_mm_any_lt(rsq20,rcutoff2))
1085             {
1086
1087             r20              = _mm_mul_ps(rsq20,rinv20);
1088
1089             /* Compute parameters for interactions between i and j atoms */
1090             qq20             = _mm_mul_ps(iq2,jq0);
1091
1092             /* EWALD ELECTROSTATICS */
1093
1094             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1095             ewrt             = _mm_mul_ps(r20,ewtabscale);
1096             ewitab           = _mm_cvttps_epi32(ewrt);
1097             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1098             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1099                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1100                                          &ewtabF,&ewtabFn);
1101             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1102             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1103
1104             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1105
1106             fscal            = felec;
1107
1108             fscal            = _mm_and_ps(fscal,cutoff_mask);
1109
1110             /* Calculate temporary vectorial force */
1111             tx               = _mm_mul_ps(fscal,dx20);
1112             ty               = _mm_mul_ps(fscal,dy20);
1113             tz               = _mm_mul_ps(fscal,dz20);
1114
1115             /* Update vectorial force */
1116             fix2             = _mm_add_ps(fix2,tx);
1117             fiy2             = _mm_add_ps(fiy2,ty);
1118             fiz2             = _mm_add_ps(fiz2,tz);
1119
1120             fjx0             = _mm_add_ps(fjx0,tx);
1121             fjy0             = _mm_add_ps(fjy0,ty);
1122             fjz0             = _mm_add_ps(fjz0,tz);
1123
1124             }
1125
1126             fjptrA             = f+j_coord_offsetA;
1127             fjptrB             = f+j_coord_offsetB;
1128             fjptrC             = f+j_coord_offsetC;
1129             fjptrD             = f+j_coord_offsetD;
1130
1131             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1132
1133             /* Inner loop uses 140 flops */
1134         }
1135
1136         if(jidx<j_index_end)
1137         {
1138
1139             /* Get j neighbor index, and coordinate index */
1140             jnrlistA         = jjnr[jidx];
1141             jnrlistB         = jjnr[jidx+1];
1142             jnrlistC         = jjnr[jidx+2];
1143             jnrlistD         = jjnr[jidx+3];
1144             /* Sign of each element will be negative for non-real atoms.
1145              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1146              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1147              */
1148             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1149             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1150             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1151             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1152             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1153             j_coord_offsetA  = DIM*jnrA;
1154             j_coord_offsetB  = DIM*jnrB;
1155             j_coord_offsetC  = DIM*jnrC;
1156             j_coord_offsetD  = DIM*jnrD;
1157
1158             /* load j atom coordinates */
1159             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1160                                               x+j_coord_offsetC,x+j_coord_offsetD,
1161                                               &jx0,&jy0,&jz0);
1162
1163             /* Calculate displacement vector */
1164             dx00             = _mm_sub_ps(ix0,jx0);
1165             dy00             = _mm_sub_ps(iy0,jy0);
1166             dz00             = _mm_sub_ps(iz0,jz0);
1167             dx10             = _mm_sub_ps(ix1,jx0);
1168             dy10             = _mm_sub_ps(iy1,jy0);
1169             dz10             = _mm_sub_ps(iz1,jz0);
1170             dx20             = _mm_sub_ps(ix2,jx0);
1171             dy20             = _mm_sub_ps(iy2,jy0);
1172             dz20             = _mm_sub_ps(iz2,jz0);
1173
1174             /* Calculate squared distance and things based on it */
1175             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1176             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1177             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1178
1179             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1180             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1181             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1182
1183             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1184             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1185             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1186
1187             /* Load parameters for j particles */
1188             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1189                                                               charge+jnrC+0,charge+jnrD+0);
1190             vdwjidx0A        = 2*vdwtype[jnrA+0];
1191             vdwjidx0B        = 2*vdwtype[jnrB+0];
1192             vdwjidx0C        = 2*vdwtype[jnrC+0];
1193             vdwjidx0D        = 2*vdwtype[jnrD+0];
1194
1195             fjx0             = _mm_setzero_ps();
1196             fjy0             = _mm_setzero_ps();
1197             fjz0             = _mm_setzero_ps();
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             if (gmx_mm_any_lt(rsq00,rcutoff2))
1204             {
1205
1206             r00              = _mm_mul_ps(rsq00,rinv00);
1207             r00              = _mm_andnot_ps(dummy_mask,r00);
1208
1209             /* Compute parameters for interactions between i and j atoms */
1210             qq00             = _mm_mul_ps(iq0,jq0);
1211             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1212                                          vdwparam+vdwioffset0+vdwjidx0B,
1213                                          vdwparam+vdwioffset0+vdwjidx0C,
1214                                          vdwparam+vdwioffset0+vdwjidx0D,
1215                                          &c6_00,&c12_00);
1216             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1217                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1218                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1219                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1220
1221             /* EWALD ELECTROSTATICS */
1222
1223             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1224             ewrt             = _mm_mul_ps(r00,ewtabscale);
1225             ewitab           = _mm_cvttps_epi32(ewrt);
1226             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1227             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1228                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1229                                          &ewtabF,&ewtabFn);
1230             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1231             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1232
1233             /* Analytical LJ-PME */
1234             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1235             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1236             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1237             exponent         = gmx_simd_exp_r(ewcljrsq);
1238             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1239             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1240             /* f6A = 6 * C6grid * (1 - poly) */
1241             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1242             /* f6B = C6grid * exponent * beta^6 */
1243             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1244             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1245             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);
1246
1247             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1248
1249             fscal            = _mm_add_ps(felec,fvdw);
1250
1251             fscal            = _mm_and_ps(fscal,cutoff_mask);
1252
1253             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1254
1255             /* Calculate temporary vectorial force */
1256             tx               = _mm_mul_ps(fscal,dx00);
1257             ty               = _mm_mul_ps(fscal,dy00);
1258             tz               = _mm_mul_ps(fscal,dz00);
1259
1260             /* Update vectorial force */
1261             fix0             = _mm_add_ps(fix0,tx);
1262             fiy0             = _mm_add_ps(fiy0,ty);
1263             fiz0             = _mm_add_ps(fiz0,tz);
1264
1265             fjx0             = _mm_add_ps(fjx0,tx);
1266             fjy0             = _mm_add_ps(fjy0,ty);
1267             fjz0             = _mm_add_ps(fjz0,tz);
1268
1269             }
1270
1271             /**************************
1272              * CALCULATE INTERACTIONS *
1273              **************************/
1274
1275             if (gmx_mm_any_lt(rsq10,rcutoff2))
1276             {
1277
1278             r10              = _mm_mul_ps(rsq10,rinv10);
1279             r10              = _mm_andnot_ps(dummy_mask,r10);
1280
1281             /* Compute parameters for interactions between i and j atoms */
1282             qq10             = _mm_mul_ps(iq1,jq0);
1283
1284             /* EWALD ELECTROSTATICS */
1285
1286             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1287             ewrt             = _mm_mul_ps(r10,ewtabscale);
1288             ewitab           = _mm_cvttps_epi32(ewrt);
1289             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1290             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1291                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1292                                          &ewtabF,&ewtabFn);
1293             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1294             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1295
1296             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1297
1298             fscal            = felec;
1299
1300             fscal            = _mm_and_ps(fscal,cutoff_mask);
1301
1302             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1303
1304             /* Calculate temporary vectorial force */
1305             tx               = _mm_mul_ps(fscal,dx10);
1306             ty               = _mm_mul_ps(fscal,dy10);
1307             tz               = _mm_mul_ps(fscal,dz10);
1308
1309             /* Update vectorial force */
1310             fix1             = _mm_add_ps(fix1,tx);
1311             fiy1             = _mm_add_ps(fiy1,ty);
1312             fiz1             = _mm_add_ps(fiz1,tz);
1313
1314             fjx0             = _mm_add_ps(fjx0,tx);
1315             fjy0             = _mm_add_ps(fjy0,ty);
1316             fjz0             = _mm_add_ps(fjz0,tz);
1317
1318             }
1319
1320             /**************************
1321              * CALCULATE INTERACTIONS *
1322              **************************/
1323
1324             if (gmx_mm_any_lt(rsq20,rcutoff2))
1325             {
1326
1327             r20              = _mm_mul_ps(rsq20,rinv20);
1328             r20              = _mm_andnot_ps(dummy_mask,r20);
1329
1330             /* Compute parameters for interactions between i and j atoms */
1331             qq20             = _mm_mul_ps(iq2,jq0);
1332
1333             /* EWALD ELECTROSTATICS */
1334
1335             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1336             ewrt             = _mm_mul_ps(r20,ewtabscale);
1337             ewitab           = _mm_cvttps_epi32(ewrt);
1338             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1339             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1340                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1341                                          &ewtabF,&ewtabFn);
1342             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1343             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1344
1345             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1346
1347             fscal            = felec;
1348
1349             fscal            = _mm_and_ps(fscal,cutoff_mask);
1350
1351             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1352
1353             /* Calculate temporary vectorial force */
1354             tx               = _mm_mul_ps(fscal,dx20);
1355             ty               = _mm_mul_ps(fscal,dy20);
1356             tz               = _mm_mul_ps(fscal,dz20);
1357
1358             /* Update vectorial force */
1359             fix2             = _mm_add_ps(fix2,tx);
1360             fiy2             = _mm_add_ps(fiy2,ty);
1361             fiz2             = _mm_add_ps(fiz2,tz);
1362
1363             fjx0             = _mm_add_ps(fjx0,tx);
1364             fjy0             = _mm_add_ps(fjy0,ty);
1365             fjz0             = _mm_add_ps(fjz0,tz);
1366
1367             }
1368
1369             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1370             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1371             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1372             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1373
1374             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1375
1376             /* Inner loop uses 143 flops */
1377         }
1378
1379         /* End of innermost loop */
1380
1381         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1382                                               f+i_coord_offset,fshift+i_shift_offset);
1383
1384         /* Increment number of inner iterations */
1385         inneriter                  += j_index_end - j_index_start;
1386
1387         /* Outer loop uses 18 flops */
1388     }
1389
1390     /* Increment number of outer iterations */
1391     outeriter        += nri;
1392
1393     /* Update outer/inner flops */
1394
1395     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*143);
1396 }