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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128           c6grid_00;
 105     __m128           c6grid_10;
 106     __m128           c6grid_20;
 107     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 108     real             *vdwgridparam;
 109     __m128           one_half  = _mm_set1_ps(0.5);
 110     __m128           minus_one = _mm_set1_ps(-1.0);
 111     __m128i          ewitab;
 112     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     real             *ewtab;
 114     __m128           dummy_mask,cutoff_mask;
 115     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 116     __m128           one     = _mm_set1_ps(1.0);
 117     __m128           two     = _mm_set1_ps(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm_set1_ps(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134     vdwgridparam     = fr->ljpme_c6grid;
 135     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 136     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 137     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 138
 139     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 140     ewtab            = fr->ic->tabq_coul_FDV0;
 141     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 142     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 147     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 148     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 149     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 152     rcutoff_scalar   = fr->rcoulomb;
 153     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 154     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 155
 156     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 157     rvdw             = _mm_set1_ps(fr->rvdw);
 158
 159     /* Avoid stupid compiler warnings */
 160     jnrA = jnrB = jnrC = jnrD = 0;
 161     j_coord_offsetA = 0;
 162     j_coord_offsetB = 0;
 163     j_coord_offsetC = 0;
 164     j_coord_offsetD = 0;
 165
 166     outeriter        = 0;
 167     inneriter        = 0;
 168
 169     for(iidx=0;iidx<4*DIM;iidx++)
 170     {
 171         scratch[iidx] = 0.0;
 172     }
 173
 174     /* Start outer loop over neighborlists */
 175     for(iidx=0; iidx<nri; iidx++)
 176     {
 177         /* Load shift vector for this list */
 178         i_shift_offset   = DIM*shiftidx[iidx];
 179
 180         /* Load limits for loop over neighbors */
 181         j_index_start    = jindex[iidx];
 182         j_index_end      = jindex[iidx+1];
 183
 184         /* Get outer coordinate index */
 185         inr              = iinr[iidx];
 186         i_coord_offset   = DIM*inr;
 187
 188         /* Load i particle coords and add shift vector */
 189         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 190                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 191
 192         fix0             = _mm_setzero_ps();
 193         fiy0             = _mm_setzero_ps();
 194         fiz0             = _mm_setzero_ps();
 195         fix1             = _mm_setzero_ps();
 196         fiy1             = _mm_setzero_ps();
 197         fiz1             = _mm_setzero_ps();
 198         fix2             = _mm_setzero_ps();
 199         fiy2             = _mm_setzero_ps();
 200         fiz2             = _mm_setzero_ps();
 201
 202         /* Reset potential sums */
 203         velecsum         = _mm_setzero_ps();
 204         vvdwsum          = _mm_setzero_ps();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             jnrC             = jjnr[jidx+2];
 214             jnrD             = jjnr[jidx+3];
 215             j_coord_offsetA  = DIM*jnrA;
 216             j_coord_offsetB  = DIM*jnrB;
 217             j_coord_offsetC  = DIM*jnrC;
 218             j_coord_offsetD  = DIM*jnrD;
 219
 220             /* load j atom coordinates */
 221             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                               x+j_coord_offsetC,x+j_coord_offsetD,
 223                                               &jx0,&jy0,&jz0);
 224
 225             /* Calculate displacement vector */
 226             dx00             = _mm_sub_ps(ix0,jx0);
 227             dy00             = _mm_sub_ps(iy0,jy0);
 228             dz00             = _mm_sub_ps(iz0,jz0);
 229             dx10             = _mm_sub_ps(ix1,jx0);
 230             dy10             = _mm_sub_ps(iy1,jy0);
 231             dz10             = _mm_sub_ps(iz1,jz0);
 232             dx20             = _mm_sub_ps(ix2,jx0);
 233             dy20             = _mm_sub_ps(iy2,jy0);
 234             dz20             = _mm_sub_ps(iz2,jz0);
 235
 236             /* Calculate squared distance and things based on it */
 237             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 238             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 239             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 240
 241             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 242             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 243             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 244
 245             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 246             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 247             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 248
 249             /* Load parameters for j particles */
 250             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 251                                                               charge+jnrC+0,charge+jnrD+0);
 252             vdwjidx0A        = 2*vdwtype[jnrA+0];
 253             vdwjidx0B        = 2*vdwtype[jnrB+0];
 254             vdwjidx0C        = 2*vdwtype[jnrC+0];
 255             vdwjidx0D        = 2*vdwtype[jnrD+0];
 256
 257             fjx0             = _mm_setzero_ps();
 258             fjy0             = _mm_setzero_ps();
 259             fjz0             = _mm_setzero_ps();
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             if (gmx_mm_any_lt(rsq00,rcutoff2))
 266             {
 267
 268             r00              = _mm_mul_ps(rsq00,rinv00);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq00             = _mm_mul_ps(iq0,jq0);
 272             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 273                                          vdwparam+vdwioffset0+vdwjidx0B,
 274                                          vdwparam+vdwioffset0+vdwjidx0C,
 275                                          vdwparam+vdwioffset0+vdwjidx0D,
 276                                          &c6_00,&c12_00);
 277
 278             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 279                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 280                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 281                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 282
 283             /* EWALD ELECTROSTATICS */
 284
 285             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 286             ewrt             = _mm_mul_ps(r00,ewtabscale);
 287             ewitab           = _mm_cvttps_epi32(ewrt);
 288             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 289             ewitab           = _mm_slli_epi32(ewitab,2);
 290             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 291             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 292             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 293             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 294             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 295             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 296             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 297             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 298             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 299
 300             /* Analytical LJ-PME */
 301             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 302             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 303             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 304             exponent         = gmx_simd_exp_r(ewcljrsq);
 305             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 306             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 307             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 308             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 309             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 310             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),
 311                                _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));
 312             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 313             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);
 314
 315             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velec            = _mm_and_ps(velec,cutoff_mask);
 319             velecsum         = _mm_add_ps(velecsum,velec);
 320             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 321             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 322
 323             fscal            = _mm_add_ps(felec,fvdw);
 324
 325             fscal            = _mm_and_ps(fscal,cutoff_mask);
 326
 327             /* Calculate temporary vectorial force */
 328             tx               = _mm_mul_ps(fscal,dx00);
 329             ty               = _mm_mul_ps(fscal,dy00);
 330             tz               = _mm_mul_ps(fscal,dz00);
 331
 332             /* Update vectorial force */
 333             fix0             = _mm_add_ps(fix0,tx);
 334             fiy0             = _mm_add_ps(fiy0,ty);
 335             fiz0             = _mm_add_ps(fiz0,tz);
 336
 337             fjx0             = _mm_add_ps(fjx0,tx);
 338             fjy0             = _mm_add_ps(fjy0,ty);
 339             fjz0             = _mm_add_ps(fjz0,tz);
 340
 341             }
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             if (gmx_mm_any_lt(rsq10,rcutoff2))
 348             {
 349
 350             r10              = _mm_mul_ps(rsq10,rinv10);
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq10             = _mm_mul_ps(iq1,jq0);
 354
 355             /* EWALD ELECTROSTATICS */
 356
 357             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 358             ewrt             = _mm_mul_ps(r10,ewtabscale);
 359             ewitab           = _mm_cvttps_epi32(ewrt);
 360             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 361             ewitab           = _mm_slli_epi32(ewitab,2);
 362             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 363             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 364             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 365             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 366             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 367             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 368             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 369             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 370             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 371
 372             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velec            = _mm_and_ps(velec,cutoff_mask);
 376             velecsum         = _mm_add_ps(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             fscal            = _mm_and_ps(fscal,cutoff_mask);
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm_mul_ps(fscal,dx10);
 384             ty               = _mm_mul_ps(fscal,dy10);
 385             tz               = _mm_mul_ps(fscal,dz10);
 386
 387             /* Update vectorial force */
 388             fix1             = _mm_add_ps(fix1,tx);
 389             fiy1             = _mm_add_ps(fiy1,ty);
 390             fiz1             = _mm_add_ps(fiz1,tz);
 391
 392             fjx0             = _mm_add_ps(fjx0,tx);
 393             fjy0             = _mm_add_ps(fjy0,ty);
 394             fjz0             = _mm_add_ps(fjz0,tz);
 395
 396             }
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             if (gmx_mm_any_lt(rsq20,rcutoff2))
 403             {
 404
 405             r20              = _mm_mul_ps(rsq20,rinv20);
 406
 407             /* Compute parameters for interactions between i and j atoms */
 408             qq20             = _mm_mul_ps(iq2,jq0);
 409
 410             /* EWALD ELECTROSTATICS */
 411
 412             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 413             ewrt             = _mm_mul_ps(r20,ewtabscale);
 414             ewitab           = _mm_cvttps_epi32(ewrt);
 415             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 416             ewitab           = _mm_slli_epi32(ewitab,2);
 417             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 418             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 419             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 420             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 421             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 422             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 423             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 424             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 425             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 426
 427             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velec            = _mm_and_ps(velec,cutoff_mask);
 431             velecsum         = _mm_add_ps(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435             fscal            = _mm_and_ps(fscal,cutoff_mask);
 436
 437             /* Calculate temporary vectorial force */
 438             tx               = _mm_mul_ps(fscal,dx20);
 439             ty               = _mm_mul_ps(fscal,dy20);
 440             tz               = _mm_mul_ps(fscal,dz20);
 441
 442             /* Update vectorial force */
 443             fix2             = _mm_add_ps(fix2,tx);
 444             fiy2             = _mm_add_ps(fiy2,ty);
 445             fiz2             = _mm_add_ps(fiz2,tz);
 446
 447             fjx0             = _mm_add_ps(fjx0,tx);
 448             fjy0             = _mm_add_ps(fjy0,ty);
 449             fjz0             = _mm_add_ps(fjz0,tz);
 450
 451             }
 452
 453             fjptrA             = f+j_coord_offsetA;
 454             fjptrB             = f+j_coord_offsetB;
 455             fjptrC             = f+j_coord_offsetC;
 456             fjptrD             = f+j_coord_offsetD;
 457
 458             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 459
 460             /* Inner loop uses 174 flops */
 461         }
 462
 463         if(jidx<j_index_end)
 464         {
 465
 466             /* Get j neighbor index, and coordinate index */
 467             jnrlistA         = jjnr[jidx];
 468             jnrlistB         = jjnr[jidx+1];
 469             jnrlistC         = jjnr[jidx+2];
 470             jnrlistD         = jjnr[jidx+3];
 471             /* Sign of each element will be negative for non-real atoms.
 472              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 473              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 474              */
 475             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 476             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 477             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 478             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 479             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 480             j_coord_offsetA  = DIM*jnrA;
 481             j_coord_offsetB  = DIM*jnrB;
 482             j_coord_offsetC  = DIM*jnrC;
 483             j_coord_offsetD  = DIM*jnrD;
 484
 485             /* load j atom coordinates */
 486             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 487                                               x+j_coord_offsetC,x+j_coord_offsetD,
 488                                               &jx0,&jy0,&jz0);
 489
 490             /* Calculate displacement vector */
 491             dx00             = _mm_sub_ps(ix0,jx0);
 492             dy00             = _mm_sub_ps(iy0,jy0);
 493             dz00             = _mm_sub_ps(iz0,jz0);
 494             dx10             = _mm_sub_ps(ix1,jx0);
 495             dy10             = _mm_sub_ps(iy1,jy0);
 496             dz10             = _mm_sub_ps(iz1,jz0);
 497             dx20             = _mm_sub_ps(ix2,jx0);
 498             dy20             = _mm_sub_ps(iy2,jy0);
 499             dz20             = _mm_sub_ps(iz2,jz0);
 500
 501             /* Calculate squared distance and things based on it */
 502             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 503             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 504             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 505
 506             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 507             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 508             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 509
 510             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 511             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 512             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 513
 514             /* Load parameters for j particles */
 515             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 516                                                               charge+jnrC+0,charge+jnrD+0);
 517             vdwjidx0A        = 2*vdwtype[jnrA+0];
 518             vdwjidx0B        = 2*vdwtype[jnrB+0];
 519             vdwjidx0C        = 2*vdwtype[jnrC+0];
 520             vdwjidx0D        = 2*vdwtype[jnrD+0];
 521
 522             fjx0             = _mm_setzero_ps();
 523             fjy0             = _mm_setzero_ps();
 524             fjz0             = _mm_setzero_ps();
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             if (gmx_mm_any_lt(rsq00,rcutoff2))
 531             {
 532
 533             r00              = _mm_mul_ps(rsq00,rinv00);
 534             r00              = _mm_andnot_ps(dummy_mask,r00);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq00             = _mm_mul_ps(iq0,jq0);
 538             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 539                                          vdwparam+vdwioffset0+vdwjidx0B,
 540                                          vdwparam+vdwioffset0+vdwjidx0C,
 541                                          vdwparam+vdwioffset0+vdwjidx0D,
 542                                          &c6_00,&c12_00);
 543
 544             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 545                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 546                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 547                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 548
 549             /* EWALD ELECTROSTATICS */
 550
 551             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 552             ewrt             = _mm_mul_ps(r00,ewtabscale);
 553             ewitab           = _mm_cvttps_epi32(ewrt);
 554             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 555             ewitab           = _mm_slli_epi32(ewitab,2);
 556             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 557             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 558             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 559             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 560             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 561             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 562             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 563             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 564             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 565
 566             /* Analytical LJ-PME */
 567             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 568             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 569             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 570             exponent         = gmx_simd_exp_r(ewcljrsq);
 571             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 572             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 573             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 574             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 575             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 576             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),
 577                                _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));
 578             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 579             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);
 580
 581             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 582
 583             /* Update potential sum for this i atom from the interaction with this j atom. */
 584             velec            = _mm_and_ps(velec,cutoff_mask);
 585             velec            = _mm_andnot_ps(dummy_mask,velec);
 586             velecsum         = _mm_add_ps(velecsum,velec);
 587             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 588             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 589             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 590
 591             fscal            = _mm_add_ps(felec,fvdw);
 592
 593             fscal            = _mm_and_ps(fscal,cutoff_mask);
 594
 595             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 596
 597             /* Calculate temporary vectorial force */
 598             tx               = _mm_mul_ps(fscal,dx00);
 599             ty               = _mm_mul_ps(fscal,dy00);
 600             tz               = _mm_mul_ps(fscal,dz00);
 601
 602             /* Update vectorial force */
 603             fix0             = _mm_add_ps(fix0,tx);
 604             fiy0             = _mm_add_ps(fiy0,ty);
 605             fiz0             = _mm_add_ps(fiz0,tz);
 606
 607             fjx0             = _mm_add_ps(fjx0,tx);
 608             fjy0             = _mm_add_ps(fjy0,ty);
 609             fjz0             = _mm_add_ps(fjz0,tz);
 610
 611             }
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             if (gmx_mm_any_lt(rsq10,rcutoff2))
 618             {
 619
 620             r10              = _mm_mul_ps(rsq10,rinv10);
 621             r10              = _mm_andnot_ps(dummy_mask,r10);
 622
 623             /* Compute parameters for interactions between i and j atoms */
 624             qq10             = _mm_mul_ps(iq1,jq0);
 625
 626             /* EWALD ELECTROSTATICS */
 627
 628             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 629             ewrt             = _mm_mul_ps(r10,ewtabscale);
 630             ewitab           = _mm_cvttps_epi32(ewrt);
 631             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 632             ewitab           = _mm_slli_epi32(ewitab,2);
 633             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 634             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 635             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 636             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 637             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 638             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 639             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 640             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 641             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 642
 643             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 644
 645             /* Update potential sum for this i atom from the interaction with this j atom. */
 646             velec            = _mm_and_ps(velec,cutoff_mask);
 647             velec            = _mm_andnot_ps(dummy_mask,velec);
 648             velecsum         = _mm_add_ps(velecsum,velec);
 649
 650             fscal            = felec;
 651
 652             fscal            = _mm_and_ps(fscal,cutoff_mask);
 653
 654             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 655
 656             /* Calculate temporary vectorial force */
 657             tx               = _mm_mul_ps(fscal,dx10);
 658             ty               = _mm_mul_ps(fscal,dy10);
 659             tz               = _mm_mul_ps(fscal,dz10);
 660
 661             /* Update vectorial force */
 662             fix1             = _mm_add_ps(fix1,tx);
 663             fiy1             = _mm_add_ps(fiy1,ty);
 664             fiz1             = _mm_add_ps(fiz1,tz);
 665
 666             fjx0             = _mm_add_ps(fjx0,tx);
 667             fjy0             = _mm_add_ps(fjy0,ty);
 668             fjz0             = _mm_add_ps(fjz0,tz);
 669
 670             }
 671
 672             /**************************
 673              * CALCULATE INTERACTIONS *
 674              **************************/
 675
 676             if (gmx_mm_any_lt(rsq20,rcutoff2))
 677             {
 678
 679             r20              = _mm_mul_ps(rsq20,rinv20);
 680             r20              = _mm_andnot_ps(dummy_mask,r20);
 681
 682             /* Compute parameters for interactions between i and j atoms */
 683             qq20             = _mm_mul_ps(iq2,jq0);
 684
 685             /* EWALD ELECTROSTATICS */
 686
 687             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 688             ewrt             = _mm_mul_ps(r20,ewtabscale);
 689             ewitab           = _mm_cvttps_epi32(ewrt);
 690             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 691             ewitab           = _mm_slli_epi32(ewitab,2);
 692             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 693             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 694             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 695             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 696             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 697             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 698             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 699             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 700             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 701
 702             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 703
 704             /* Update potential sum for this i atom from the interaction with this j atom. */
 705             velec            = _mm_and_ps(velec,cutoff_mask);
 706             velec            = _mm_andnot_ps(dummy_mask,velec);
 707             velecsum         = _mm_add_ps(velecsum,velec);
 708
 709             fscal            = felec;
 710
 711             fscal            = _mm_and_ps(fscal,cutoff_mask);
 712
 713             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 714
 715             /* Calculate temporary vectorial force */
 716             tx               = _mm_mul_ps(fscal,dx20);
 717             ty               = _mm_mul_ps(fscal,dy20);
 718             tz               = _mm_mul_ps(fscal,dz20);
 719
 720             /* Update vectorial force */
 721             fix2             = _mm_add_ps(fix2,tx);
 722             fiy2             = _mm_add_ps(fiy2,ty);
 723             fiz2             = _mm_add_ps(fiz2,tz);
 724
 725             fjx0             = _mm_add_ps(fjx0,tx);
 726             fjy0             = _mm_add_ps(fjy0,ty);
 727             fjz0             = _mm_add_ps(fjz0,tz);
 728
 729             }
 730
 731             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 732             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 733             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 734             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 735
 736             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 737
 738             /* Inner loop uses 177 flops */
 739         }
 740
 741         /* End of innermost loop */
 742
 743         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 744                                               f+i_coord_offset,fshift+i_shift_offset);
 745
 746         ggid                        = gid[iidx];
 747         /* Update potential energies */
 748         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 749         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 750
 751         /* Increment number of inner iterations */
 752         inneriter                  += j_index_end - j_index_start;
 753
 754         /* Outer loop uses 20 flops */
 755     }
 756
 757     /* Increment number of outer iterations */
 758     outeriter        += nri;
 759
 760     /* Update outer/inner flops */
 761
 762     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*177);
 763 }
 764 /*
 765  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse4_1_single
 766  * Electrostatics interaction: Ewald
 767  * VdW interaction:            LJEwald
 768  * Geometry:                   Water3-Particle
 769  * Calculate force/pot:        Force
 770  */
 771 void
 772 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse4_1_single
 773                     (t_nblist                    * gmx_restrict       nlist,
 774                      rvec                        * gmx_restrict          xx,
 775                      rvec                        * gmx_restrict          ff,
 776                      t_forcerec                  * gmx_restrict          fr,
 777                      t_mdatoms                   * gmx_restrict     mdatoms,
 778                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 779                      t_nrnb                      * gmx_restrict        nrnb)
 780 {
 781     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 782      * just 0 for non-waters.
 783      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 784      * jnr indices corresponding to data put in the four positions in the SIMD register.
 785      */
 786     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 787     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 788     int              jnrA,jnrB,jnrC,jnrD;
 789     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 790     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 791     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 792     real             rcutoff_scalar;
 793     real             *shiftvec,*fshift,*x,*f;
 794     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 795     real             scratch[4*DIM];
 796     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 797     int              vdwioffset0;
 798     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 799     int              vdwioffset1;
 800     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 801     int              vdwioffset2;
 802     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 803     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 804     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 805     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 806     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 807     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 808     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 809     real             *charge;
 810     int              nvdwtype;
 811     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 812     int              *vdwtype;
 813     real             *vdwparam;
 814     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 815     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 816     __m128           c6grid_00;
 817     __m128           c6grid_10;
 818     __m128           c6grid_20;
 819     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 820     real             *vdwgridparam;
 821     __m128           one_half  = _mm_set1_ps(0.5);
 822     __m128           minus_one = _mm_set1_ps(-1.0);
 823     __m128i          ewitab;
 824     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 825     real             *ewtab;
 826     __m128           dummy_mask,cutoff_mask;
 827     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 828     __m128           one     = _mm_set1_ps(1.0);
 829     __m128           two     = _mm_set1_ps(2.0);
 830     x                = xx[0];
 831     f                = ff[0];
 832
 833     nri              = nlist->nri;
 834     iinr             = nlist->iinr;
 835     jindex           = nlist->jindex;
 836     jjnr             = nlist->jjnr;
 837     shiftidx         = nlist->shift;
 838     gid              = nlist->gid;
 839     shiftvec         = fr->shift_vec[0];
 840     fshift           = fr->fshift[0];
 841     facel            = _mm_set1_ps(fr->epsfac);
 842     charge           = mdatoms->chargeA;
 843     nvdwtype         = fr->ntype;
 844     vdwparam         = fr->nbfp;
 845     vdwtype          = mdatoms->typeA;
 846     vdwgridparam     = fr->ljpme_c6grid;
 847     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 848     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 849     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 850
 851     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 852     ewtab            = fr->ic->tabq_coul_F;
 853     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 854     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 855
 856     /* Setup water-specific parameters */
 857     inr              = nlist->iinr[0];
 858     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 859     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 860     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 861     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 862
 863     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 864     rcutoff_scalar   = fr->rcoulomb;
 865     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 866     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 867
 868     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 869     rvdw             = _mm_set1_ps(fr->rvdw);
 870
 871     /* Avoid stupid compiler warnings */
 872     jnrA = jnrB = jnrC = jnrD = 0;
 873     j_coord_offsetA = 0;
 874     j_coord_offsetB = 0;
 875     j_coord_offsetC = 0;
 876     j_coord_offsetD = 0;
 877
 878     outeriter        = 0;
 879     inneriter        = 0;
 880
 881     for(iidx=0;iidx<4*DIM;iidx++)
 882     {
 883         scratch[iidx] = 0.0;
 884     }
 885
 886     /* Start outer loop over neighborlists */
 887     for(iidx=0; iidx<nri; iidx++)
 888     {
 889         /* Load shift vector for this list */
 890         i_shift_offset   = DIM*shiftidx[iidx];
 891
 892         /* Load limits for loop over neighbors */
 893         j_index_start    = jindex[iidx];
 894         j_index_end      = jindex[iidx+1];
 895
 896         /* Get outer coordinate index */
 897         inr              = iinr[iidx];
 898         i_coord_offset   = DIM*inr;
 899
 900         /* Load i particle coords and add shift vector */
 901         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 902                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 903
 904         fix0             = _mm_setzero_ps();
 905         fiy0             = _mm_setzero_ps();
 906         fiz0             = _mm_setzero_ps();
 907         fix1             = _mm_setzero_ps();
 908         fiy1             = _mm_setzero_ps();
 909         fiz1             = _mm_setzero_ps();
 910         fix2             = _mm_setzero_ps();
 911         fiy2             = _mm_setzero_ps();
 912         fiz2             = _mm_setzero_ps();
 913
 914         /* Start inner kernel loop */
 915         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 916         {
 917
 918             /* Get j neighbor index, and coordinate index */
 919             jnrA             = jjnr[jidx];
 920             jnrB             = jjnr[jidx+1];
 921             jnrC             = jjnr[jidx+2];
 922             jnrD             = jjnr[jidx+3];
 923             j_coord_offsetA  = DIM*jnrA;
 924             j_coord_offsetB  = DIM*jnrB;
 925             j_coord_offsetC  = DIM*jnrC;
 926             j_coord_offsetD  = DIM*jnrD;
 927
 928             /* load j atom coordinates */
 929             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 930                                               x+j_coord_offsetC,x+j_coord_offsetD,
 931                                               &jx0,&jy0,&jz0);
 932
 933             /* Calculate displacement vector */
 934             dx00             = _mm_sub_ps(ix0,jx0);
 935             dy00             = _mm_sub_ps(iy0,jy0);
 936             dz00             = _mm_sub_ps(iz0,jz0);
 937             dx10             = _mm_sub_ps(ix1,jx0);
 938             dy10             = _mm_sub_ps(iy1,jy0);
 939             dz10             = _mm_sub_ps(iz1,jz0);
 940             dx20             = _mm_sub_ps(ix2,jx0);
 941             dy20             = _mm_sub_ps(iy2,jy0);
 942             dz20             = _mm_sub_ps(iz2,jz0);
 943
 944             /* Calculate squared distance and things based on it */
 945             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 946             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 947             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 948
 949             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 950             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 951             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 952
 953             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 954             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 955             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 956
 957             /* Load parameters for j particles */
 958             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 959                                                               charge+jnrC+0,charge+jnrD+0);
 960             vdwjidx0A        = 2*vdwtype[jnrA+0];
 961             vdwjidx0B        = 2*vdwtype[jnrB+0];
 962             vdwjidx0C        = 2*vdwtype[jnrC+0];
 963             vdwjidx0D        = 2*vdwtype[jnrD+0];
 964
 965             fjx0             = _mm_setzero_ps();
 966             fjy0             = _mm_setzero_ps();
 967             fjz0             = _mm_setzero_ps();
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             if (gmx_mm_any_lt(rsq00,rcutoff2))
 974             {
 975
 976             r00              = _mm_mul_ps(rsq00,rinv00);
 977
 978             /* Compute parameters for interactions between i and j atoms */
 979             qq00             = _mm_mul_ps(iq0,jq0);
 980             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 981                                          vdwparam+vdwioffset0+vdwjidx0B,
 982                                          vdwparam+vdwioffset0+vdwjidx0C,
 983                                          vdwparam+vdwioffset0+vdwjidx0D,
 984                                          &c6_00,&c12_00);
 985
 986             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 987                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 988                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 989                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 990
 991             /* EWALD ELECTROSTATICS */
 992
 993             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 994             ewrt             = _mm_mul_ps(r00,ewtabscale);
 995             ewitab           = _mm_cvttps_epi32(ewrt);
 996             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 997             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 998                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 999                                          &ewtabF,&ewtabFn);
1000             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1001             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1002
1003             /* Analytical LJ-PME */
1004             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1005             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1006             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1007             exponent         = gmx_simd_exp_r(ewcljrsq);
1008             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1009             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1010             /* f6A = 6 * C6grid * (1 - poly) */
1011             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1012             /* f6B = C6grid * exponent * beta^6 */
1013             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1014             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1015             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);
1016
1017             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1018
1019             fscal            = _mm_add_ps(felec,fvdw);
1020
1021             fscal            = _mm_and_ps(fscal,cutoff_mask);
1022
1023             /* Calculate temporary vectorial force */
1024             tx               = _mm_mul_ps(fscal,dx00);
1025             ty               = _mm_mul_ps(fscal,dy00);
1026             tz               = _mm_mul_ps(fscal,dz00);
1027
1028             /* Update vectorial force */
1029             fix0             = _mm_add_ps(fix0,tx);
1030             fiy0             = _mm_add_ps(fiy0,ty);
1031             fiz0             = _mm_add_ps(fiz0,tz);
1032
1033             fjx0             = _mm_add_ps(fjx0,tx);
1034             fjy0             = _mm_add_ps(fjy0,ty);
1035             fjz0             = _mm_add_ps(fjz0,tz);
1036
1037             }
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             if (gmx_mm_any_lt(rsq10,rcutoff2))
1044             {
1045
1046             r10              = _mm_mul_ps(rsq10,rinv10);
1047
1048             /* Compute parameters for interactions between i and j atoms */
1049             qq10             = _mm_mul_ps(iq1,jq0);
1050
1051             /* EWALD ELECTROSTATICS */
1052
1053             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1054             ewrt             = _mm_mul_ps(r10,ewtabscale);
1055             ewitab           = _mm_cvttps_epi32(ewrt);
1056             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1057             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1058                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1059                                          &ewtabF,&ewtabFn);
1060             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1061             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1062
1063             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1064
1065             fscal            = felec;
1066
1067             fscal            = _mm_and_ps(fscal,cutoff_mask);
1068
1069             /* Calculate temporary vectorial force */
1070             tx               = _mm_mul_ps(fscal,dx10);
1071             ty               = _mm_mul_ps(fscal,dy10);
1072             tz               = _mm_mul_ps(fscal,dz10);
1073
1074             /* Update vectorial force */
1075             fix1             = _mm_add_ps(fix1,tx);
1076             fiy1             = _mm_add_ps(fiy1,ty);
1077             fiz1             = _mm_add_ps(fiz1,tz);
1078
1079             fjx0             = _mm_add_ps(fjx0,tx);
1080             fjy0             = _mm_add_ps(fjy0,ty);
1081             fjz0             = _mm_add_ps(fjz0,tz);
1082
1083             }
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             if (gmx_mm_any_lt(rsq20,rcutoff2))
1090             {
1091
1092             r20              = _mm_mul_ps(rsq20,rinv20);
1093
1094             /* Compute parameters for interactions between i and j atoms */
1095             qq20             = _mm_mul_ps(iq2,jq0);
1096
1097             /* EWALD ELECTROSTATICS */
1098
1099             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1100             ewrt             = _mm_mul_ps(r20,ewtabscale);
1101             ewitab           = _mm_cvttps_epi32(ewrt);
1102             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1103             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1104                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1105                                          &ewtabF,&ewtabFn);
1106             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1107             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1108
1109             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1110
1111             fscal            = felec;
1112
1113             fscal            = _mm_and_ps(fscal,cutoff_mask);
1114
1115             /* Calculate temporary vectorial force */
1116             tx               = _mm_mul_ps(fscal,dx20);
1117             ty               = _mm_mul_ps(fscal,dy20);
1118             tz               = _mm_mul_ps(fscal,dz20);
1119
1120             /* Update vectorial force */
1121             fix2             = _mm_add_ps(fix2,tx);
1122             fiy2             = _mm_add_ps(fiy2,ty);
1123             fiz2             = _mm_add_ps(fiz2,tz);
1124
1125             fjx0             = _mm_add_ps(fjx0,tx);
1126             fjy0             = _mm_add_ps(fjy0,ty);
1127             fjz0             = _mm_add_ps(fjz0,tz);
1128
1129             }
1130
1131             fjptrA             = f+j_coord_offsetA;
1132             fjptrB             = f+j_coord_offsetB;
1133             fjptrC             = f+j_coord_offsetC;
1134             fjptrD             = f+j_coord_offsetD;
1135
1136             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1137
1138             /* Inner loop uses 140 flops */
1139         }
1140
1141         if(jidx<j_index_end)
1142         {
1143
1144             /* Get j neighbor index, and coordinate index */
1145             jnrlistA         = jjnr[jidx];
1146             jnrlistB         = jjnr[jidx+1];
1147             jnrlistC         = jjnr[jidx+2];
1148             jnrlistD         = jjnr[jidx+3];
1149             /* Sign of each element will be negative for non-real atoms.
1150              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1151              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1152              */
1153             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1154             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1155             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1156             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1157             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1158             j_coord_offsetA  = DIM*jnrA;
1159             j_coord_offsetB  = DIM*jnrB;
1160             j_coord_offsetC  = DIM*jnrC;
1161             j_coord_offsetD  = DIM*jnrD;
1162
1163             /* load j atom coordinates */
1164             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1165                                               x+j_coord_offsetC,x+j_coord_offsetD,
1166                                               &jx0,&jy0,&jz0);
1167
1168             /* Calculate displacement vector */
1169             dx00             = _mm_sub_ps(ix0,jx0);
1170             dy00             = _mm_sub_ps(iy0,jy0);
1171             dz00             = _mm_sub_ps(iz0,jz0);
1172             dx10             = _mm_sub_ps(ix1,jx0);
1173             dy10             = _mm_sub_ps(iy1,jy0);
1174             dz10             = _mm_sub_ps(iz1,jz0);
1175             dx20             = _mm_sub_ps(ix2,jx0);
1176             dy20             = _mm_sub_ps(iy2,jy0);
1177             dz20             = _mm_sub_ps(iz2,jz0);
1178
1179             /* Calculate squared distance and things based on it */
1180             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1181             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1182             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1183
1184             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1185             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1186             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1187
1188             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1189             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1190             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1191
1192             /* Load parameters for j particles */
1193             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1194                                                               charge+jnrC+0,charge+jnrD+0);
1195             vdwjidx0A        = 2*vdwtype[jnrA+0];
1196             vdwjidx0B        = 2*vdwtype[jnrB+0];
1197             vdwjidx0C        = 2*vdwtype[jnrC+0];
1198             vdwjidx0D        = 2*vdwtype[jnrD+0];
1199
1200             fjx0             = _mm_setzero_ps();
1201             fjy0             = _mm_setzero_ps();
1202             fjz0             = _mm_setzero_ps();
1203
1204             /**************************
1205              * CALCULATE INTERACTIONS *
1206              **************************/
1207
1208             if (gmx_mm_any_lt(rsq00,rcutoff2))
1209             {
1210
1211             r00              = _mm_mul_ps(rsq00,rinv00);
1212             r00              = _mm_andnot_ps(dummy_mask,r00);
1213
1214             /* Compute parameters for interactions between i and j atoms */
1215             qq00             = _mm_mul_ps(iq0,jq0);
1216             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1217                                          vdwparam+vdwioffset0+vdwjidx0B,
1218                                          vdwparam+vdwioffset0+vdwjidx0C,
1219                                          vdwparam+vdwioffset0+vdwjidx0D,
1220                                          &c6_00,&c12_00);
1221
1222             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1223                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1224                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1225                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1226
1227             /* EWALD ELECTROSTATICS */
1228
1229             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1230             ewrt             = _mm_mul_ps(r00,ewtabscale);
1231             ewitab           = _mm_cvttps_epi32(ewrt);
1232             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1233             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1234                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1235                                          &ewtabF,&ewtabFn);
1236             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1237             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1238
1239             /* Analytical LJ-PME */
1240             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1241             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1242             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1243             exponent         = gmx_simd_exp_r(ewcljrsq);
1244             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1245             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1246             /* f6A = 6 * C6grid * (1 - poly) */
1247             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1248             /* f6B = C6grid * exponent * beta^6 */
1249             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1250             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1251             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);
1252
1253             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1254
1255             fscal            = _mm_add_ps(felec,fvdw);
1256
1257             fscal            = _mm_and_ps(fscal,cutoff_mask);
1258
1259             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1260
1261             /* Calculate temporary vectorial force */
1262             tx               = _mm_mul_ps(fscal,dx00);
1263             ty               = _mm_mul_ps(fscal,dy00);
1264             tz               = _mm_mul_ps(fscal,dz00);
1265
1266             /* Update vectorial force */
1267             fix0             = _mm_add_ps(fix0,tx);
1268             fiy0             = _mm_add_ps(fiy0,ty);
1269             fiz0             = _mm_add_ps(fiz0,tz);
1270
1271             fjx0             = _mm_add_ps(fjx0,tx);
1272             fjy0             = _mm_add_ps(fjy0,ty);
1273             fjz0             = _mm_add_ps(fjz0,tz);
1274
1275             }
1276
1277             /**************************
1278              * CALCULATE INTERACTIONS *
1279              **************************/
1280
1281             if (gmx_mm_any_lt(rsq10,rcutoff2))
1282             {
1283
1284             r10              = _mm_mul_ps(rsq10,rinv10);
1285             r10              = _mm_andnot_ps(dummy_mask,r10);
1286
1287             /* Compute parameters for interactions between i and j atoms */
1288             qq10             = _mm_mul_ps(iq1,jq0);
1289
1290             /* EWALD ELECTROSTATICS */
1291
1292             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1293             ewrt             = _mm_mul_ps(r10,ewtabscale);
1294             ewitab           = _mm_cvttps_epi32(ewrt);
1295             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1296             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1297                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1298                                          &ewtabF,&ewtabFn);
1299             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1300             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1301
1302             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1303
1304             fscal            = felec;
1305
1306             fscal            = _mm_and_ps(fscal,cutoff_mask);
1307
1308             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1309
1310             /* Calculate temporary vectorial force */
1311             tx               = _mm_mul_ps(fscal,dx10);
1312             ty               = _mm_mul_ps(fscal,dy10);
1313             tz               = _mm_mul_ps(fscal,dz10);
1314
1315             /* Update vectorial force */
1316             fix1             = _mm_add_ps(fix1,tx);
1317             fiy1             = _mm_add_ps(fiy1,ty);
1318             fiz1             = _mm_add_ps(fiz1,tz);
1319
1320             fjx0             = _mm_add_ps(fjx0,tx);
1321             fjy0             = _mm_add_ps(fjy0,ty);
1322             fjz0             = _mm_add_ps(fjz0,tz);
1323
1324             }
1325
1326             /**************************
1327              * CALCULATE INTERACTIONS *
1328              **************************/
1329
1330             if (gmx_mm_any_lt(rsq20,rcutoff2))
1331             {
1332
1333             r20              = _mm_mul_ps(rsq20,rinv20);
1334             r20              = _mm_andnot_ps(dummy_mask,r20);
1335
1336             /* Compute parameters for interactions between i and j atoms */
1337             qq20             = _mm_mul_ps(iq2,jq0);
1338
1339             /* EWALD ELECTROSTATICS */
1340
1341             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1342             ewrt             = _mm_mul_ps(r20,ewtabscale);
1343             ewitab           = _mm_cvttps_epi32(ewrt);
1344             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1345             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1346                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1347                                          &ewtabF,&ewtabFn);
1348             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1349             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1350
1351             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1352
1353             fscal            = felec;
1354
1355             fscal            = _mm_and_ps(fscal,cutoff_mask);
1356
1357             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1358
1359             /* Calculate temporary vectorial force */
1360             tx               = _mm_mul_ps(fscal,dx20);
1361             ty               = _mm_mul_ps(fscal,dy20);
1362             tz               = _mm_mul_ps(fscal,dz20);
1363
1364             /* Update vectorial force */
1365             fix2             = _mm_add_ps(fix2,tx);
1366             fiy2             = _mm_add_ps(fiy2,ty);
1367             fiz2             = _mm_add_ps(fiz2,tz);
1368
1369             fjx0             = _mm_add_ps(fjx0,tx);
1370             fjy0             = _mm_add_ps(fjy0,ty);
1371             fjz0             = _mm_add_ps(fjz0,tz);
1372
1373             }
1374
1375             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1376             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1377             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1378             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1379
1380             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1381
1382             /* Inner loop uses 143 flops */
1383         }
1384
1385         /* End of innermost loop */
1386
1387         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1388                                               f+i_coord_offset,fshift+i_shift_offset);
1389
1390         /* Increment number of inner iterations */
1391         inneriter                  += j_index_end - j_index_start;
1392
1393         /* Outer loop uses 18 flops */
1394     }
1395
1396     /* Increment number of outer iterations */
1397     outeriter        += nri;
1398
1399     /* Update outer/inner flops */
1400
1401     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*143);
1402 }