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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_sse4_1_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_sse4_1_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
 276             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 277                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 278                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 279                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 280
 281             /* EWALD ELECTROSTATICS */
 282
 283             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 284             ewrt             = _mm_mul_ps(r00,ewtabscale);
 285             ewitab           = _mm_cvttps_epi32(ewrt);
 286             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 287             ewitab           = _mm_slli_epi32(ewitab,2);
 288             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 289             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 290             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 291             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 292             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 293             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 294             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 295             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 296             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 297
 298             /* Analytical LJ-PME */
 299             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 300             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 301             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 302             exponent         = gmx_simd_exp_r(ewcljrsq);
 303             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 304             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 305             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 306             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 307             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 308             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),
 309                                _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));
 310             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 311             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);
 312
 313             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velec            = _mm_and_ps(velec,cutoff_mask);
 317             velecsum         = _mm_add_ps(velecsum,velec);
 318             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 319             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 320
 321             fscal            = _mm_add_ps(felec,fvdw);
 322
 323             fscal            = _mm_and_ps(fscal,cutoff_mask);
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm_mul_ps(fscal,dx00);
 327             ty               = _mm_mul_ps(fscal,dy00);
 328             tz               = _mm_mul_ps(fscal,dz00);
 329
 330             /* Update vectorial force */
 331             fix0             = _mm_add_ps(fix0,tx);
 332             fiy0             = _mm_add_ps(fiy0,ty);
 333             fiz0             = _mm_add_ps(fiz0,tz);
 334
 335             fjx0             = _mm_add_ps(fjx0,tx);
 336             fjy0             = _mm_add_ps(fjy0,ty);
 337             fjz0             = _mm_add_ps(fjz0,tz);
 338
 339             }
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             if (gmx_mm_any_lt(rsq10,rcutoff2))
 346             {
 347
 348             r10              = _mm_mul_ps(rsq10,rinv10);
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq10             = _mm_mul_ps(iq1,jq0);
 352
 353             /* EWALD ELECTROSTATICS */
 354
 355             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 356             ewrt             = _mm_mul_ps(r10,ewtabscale);
 357             ewitab           = _mm_cvttps_epi32(ewrt);
 358             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 359             ewitab           = _mm_slli_epi32(ewitab,2);
 360             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 361             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 362             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 363             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 364             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 365             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 366             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 367             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 368             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 369
 370             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velec            = _mm_and_ps(velec,cutoff_mask);
 374             velecsum         = _mm_add_ps(velecsum,velec);
 375
 376             fscal            = felec;
 377
 378             fscal            = _mm_and_ps(fscal,cutoff_mask);
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm_mul_ps(fscal,dx10);
 382             ty               = _mm_mul_ps(fscal,dy10);
 383             tz               = _mm_mul_ps(fscal,dz10);
 384
 385             /* Update vectorial force */
 386             fix1             = _mm_add_ps(fix1,tx);
 387             fiy1             = _mm_add_ps(fiy1,ty);
 388             fiz1             = _mm_add_ps(fiz1,tz);
 389
 390             fjx0             = _mm_add_ps(fjx0,tx);
 391             fjy0             = _mm_add_ps(fjy0,ty);
 392             fjz0             = _mm_add_ps(fjz0,tz);
 393
 394             }
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             if (gmx_mm_any_lt(rsq20,rcutoff2))
 401             {
 402
 403             r20              = _mm_mul_ps(rsq20,rinv20);
 404
 405             /* Compute parameters for interactions between i and j atoms */
 406             qq20             = _mm_mul_ps(iq2,jq0);
 407
 408             /* EWALD ELECTROSTATICS */
 409
 410             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 411             ewrt             = _mm_mul_ps(r20,ewtabscale);
 412             ewitab           = _mm_cvttps_epi32(ewrt);
 413             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 414             ewitab           = _mm_slli_epi32(ewitab,2);
 415             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 416             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 417             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 418             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 419             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 420             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 421             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 422             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 423             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 424
 425             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 426
 427             /* Update potential sum for this i atom from the interaction with this j atom. */
 428             velec            = _mm_and_ps(velec,cutoff_mask);
 429             velecsum         = _mm_add_ps(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             fscal            = _mm_and_ps(fscal,cutoff_mask);
 434
 435             /* Calculate temporary vectorial force */
 436             tx               = _mm_mul_ps(fscal,dx20);
 437             ty               = _mm_mul_ps(fscal,dy20);
 438             tz               = _mm_mul_ps(fscal,dz20);
 439
 440             /* Update vectorial force */
 441             fix2             = _mm_add_ps(fix2,tx);
 442             fiy2             = _mm_add_ps(fiy2,ty);
 443             fiz2             = _mm_add_ps(fiz2,tz);
 444
 445             fjx0             = _mm_add_ps(fjx0,tx);
 446             fjy0             = _mm_add_ps(fjy0,ty);
 447             fjz0             = _mm_add_ps(fjz0,tz);
 448
 449             }
 450
 451             fjptrA             = f+j_coord_offsetA;
 452             fjptrB             = f+j_coord_offsetB;
 453             fjptrC             = f+j_coord_offsetC;
 454             fjptrD             = f+j_coord_offsetD;
 455
 456             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 457
 458             /* Inner loop uses 174 flops */
 459         }
 460
 461         if(jidx<j_index_end)
 462         {
 463
 464             /* Get j neighbor index, and coordinate index */
 465             jnrlistA         = jjnr[jidx];
 466             jnrlistB         = jjnr[jidx+1];
 467             jnrlistC         = jjnr[jidx+2];
 468             jnrlistD         = jjnr[jidx+3];
 469             /* Sign of each element will be negative for non-real atoms.
 470              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 471              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 472              */
 473             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 474             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 475             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 476             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 477             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 478             j_coord_offsetA  = DIM*jnrA;
 479             j_coord_offsetB  = DIM*jnrB;
 480             j_coord_offsetC  = DIM*jnrC;
 481             j_coord_offsetD  = DIM*jnrD;
 482
 483             /* load j atom coordinates */
 484             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 485                                               x+j_coord_offsetC,x+j_coord_offsetD,
 486                                               &jx0,&jy0,&jz0);
 487
 488             /* Calculate displacement vector */
 489             dx00             = _mm_sub_ps(ix0,jx0);
 490             dy00             = _mm_sub_ps(iy0,jy0);
 491             dz00             = _mm_sub_ps(iz0,jz0);
 492             dx10             = _mm_sub_ps(ix1,jx0);
 493             dy10             = _mm_sub_ps(iy1,jy0);
 494             dz10             = _mm_sub_ps(iz1,jz0);
 495             dx20             = _mm_sub_ps(ix2,jx0);
 496             dy20             = _mm_sub_ps(iy2,jy0);
 497             dz20             = _mm_sub_ps(iz2,jz0);
 498
 499             /* Calculate squared distance and things based on it */
 500             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 501             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 502             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 503
 504             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 505             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 506             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 507
 508             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 509             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 510             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 511
 512             /* Load parameters for j particles */
 513             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 514                                                               charge+jnrC+0,charge+jnrD+0);
 515             vdwjidx0A        = 2*vdwtype[jnrA+0];
 516             vdwjidx0B        = 2*vdwtype[jnrB+0];
 517             vdwjidx0C        = 2*vdwtype[jnrC+0];
 518             vdwjidx0D        = 2*vdwtype[jnrD+0];
 519
 520             fjx0             = _mm_setzero_ps();
 521             fjy0             = _mm_setzero_ps();
 522             fjz0             = _mm_setzero_ps();
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             if (gmx_mm_any_lt(rsq00,rcutoff2))
 529             {
 530
 531             r00              = _mm_mul_ps(rsq00,rinv00);
 532             r00              = _mm_andnot_ps(dummy_mask,r00);
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq00             = _mm_mul_ps(iq0,jq0);
 536             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 537                                          vdwparam+vdwioffset0+vdwjidx0B,
 538                                          vdwparam+vdwioffset0+vdwjidx0C,
 539                                          vdwparam+vdwioffset0+vdwjidx0D,
 540                                          &c6_00,&c12_00);
 541
 542             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 543                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 544                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 545                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 546
 547             /* EWALD ELECTROSTATICS */
 548
 549             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 550             ewrt             = _mm_mul_ps(r00,ewtabscale);
 551             ewitab           = _mm_cvttps_epi32(ewrt);
 552             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 553             ewitab           = _mm_slli_epi32(ewitab,2);
 554             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 555             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 556             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 557             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 558             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 559             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 560             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 561             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 562             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 563
 564             /* Analytical LJ-PME */
 565             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 566             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 567             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 568             exponent         = gmx_simd_exp_r(ewcljrsq);
 569             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 570             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 571             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 572             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 573             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 574             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
 575                                _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 576             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 577             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 578
 579             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm_and_ps(velec,cutoff_mask);
 583             velec            = _mm_andnot_ps(dummy_mask,velec);
 584             velecsum         = _mm_add_ps(velecsum,velec);
 585             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 586             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 587             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 588
 589             fscal            = _mm_add_ps(felec,fvdw);
 590
 591             fscal            = _mm_and_ps(fscal,cutoff_mask);
 592
 593             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm_mul_ps(fscal,dx00);
 597             ty               = _mm_mul_ps(fscal,dy00);
 598             tz               = _mm_mul_ps(fscal,dz00);
 599
 600             /* Update vectorial force */
 601             fix0             = _mm_add_ps(fix0,tx);
 602             fiy0             = _mm_add_ps(fiy0,ty);
 603             fiz0             = _mm_add_ps(fiz0,tz);
 604
 605             fjx0             = _mm_add_ps(fjx0,tx);
 606             fjy0             = _mm_add_ps(fjy0,ty);
 607             fjz0             = _mm_add_ps(fjz0,tz);
 608
 609             }
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             if (gmx_mm_any_lt(rsq10,rcutoff2))
 616             {
 617
 618             r10              = _mm_mul_ps(rsq10,rinv10);
 619             r10              = _mm_andnot_ps(dummy_mask,r10);
 620
 621             /* Compute parameters for interactions between i and j atoms */
 622             qq10             = _mm_mul_ps(iq1,jq0);
 623
 624             /* EWALD ELECTROSTATICS */
 625
 626             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 627             ewrt             = _mm_mul_ps(r10,ewtabscale);
 628             ewitab           = _mm_cvttps_epi32(ewrt);
 629             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 630             ewitab           = _mm_slli_epi32(ewitab,2);
 631             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 632             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 633             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 634             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 635             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 636             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 637             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 638             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 639             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 640
 641             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 642
 643             /* Update potential sum for this i atom from the interaction with this j atom. */
 644             velec            = _mm_and_ps(velec,cutoff_mask);
 645             velec            = _mm_andnot_ps(dummy_mask,velec);
 646             velecsum         = _mm_add_ps(velecsum,velec);
 647
 648             fscal            = felec;
 649
 650             fscal            = _mm_and_ps(fscal,cutoff_mask);
 651
 652             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 653
 654             /* Calculate temporary vectorial force */
 655             tx               = _mm_mul_ps(fscal,dx10);
 656             ty               = _mm_mul_ps(fscal,dy10);
 657             tz               = _mm_mul_ps(fscal,dz10);
 658
 659             /* Update vectorial force */
 660             fix1             = _mm_add_ps(fix1,tx);
 661             fiy1             = _mm_add_ps(fiy1,ty);
 662             fiz1             = _mm_add_ps(fiz1,tz);
 663
 664             fjx0             = _mm_add_ps(fjx0,tx);
 665             fjy0             = _mm_add_ps(fjy0,ty);
 666             fjz0             = _mm_add_ps(fjz0,tz);
 667
 668             }
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             if (gmx_mm_any_lt(rsq20,rcutoff2))
 675             {
 676
 677             r20              = _mm_mul_ps(rsq20,rinv20);
 678             r20              = _mm_andnot_ps(dummy_mask,r20);
 679
 680             /* Compute parameters for interactions between i and j atoms */
 681             qq20             = _mm_mul_ps(iq2,jq0);
 682
 683             /* EWALD ELECTROSTATICS */
 684
 685             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 686             ewrt             = _mm_mul_ps(r20,ewtabscale);
 687             ewitab           = _mm_cvttps_epi32(ewrt);
 688             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 689             ewitab           = _mm_slli_epi32(ewitab,2);
 690             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 691             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 692             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 693             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 694             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 695             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 696             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 697             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 698             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 699
 700             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 701
 702             /* Update potential sum for this i atom from the interaction with this j atom. */
 703             velec            = _mm_and_ps(velec,cutoff_mask);
 704             velec            = _mm_andnot_ps(dummy_mask,velec);
 705             velecsum         = _mm_add_ps(velecsum,velec);
 706
 707             fscal            = felec;
 708
 709             fscal            = _mm_and_ps(fscal,cutoff_mask);
 710
 711             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 712
 713             /* Calculate temporary vectorial force */
 714             tx               = _mm_mul_ps(fscal,dx20);
 715             ty               = _mm_mul_ps(fscal,dy20);
 716             tz               = _mm_mul_ps(fscal,dz20);
 717
 718             /* Update vectorial force */
 719             fix2             = _mm_add_ps(fix2,tx);
 720             fiy2             = _mm_add_ps(fiy2,ty);
 721             fiz2             = _mm_add_ps(fiz2,tz);
 722
 723             fjx0             = _mm_add_ps(fjx0,tx);
 724             fjy0             = _mm_add_ps(fjy0,ty);
 725             fjz0             = _mm_add_ps(fjz0,tz);
 726
 727             }
 728
 729             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 730             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 731             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 732             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 733
 734             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 735
 736             /* Inner loop uses 177 flops */
 737         }
 738
 739         /* End of innermost loop */
 740
 741         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 742                                               f+i_coord_offset,fshift+i_shift_offset);
 743
 744         ggid                        = gid[iidx];
 745         /* Update potential energies */
 746         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 747         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 748
 749         /* Increment number of inner iterations */
 750         inneriter                  += j_index_end - j_index_start;
 751
 752         /* Outer loop uses 20 flops */
 753     }
 754
 755     /* Increment number of outer iterations */
 756     outeriter        += nri;
 757
 758     /* Update outer/inner flops */
 759
 760     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*177);
 761 }
 762 /*
 763  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse4_1_single
 764  * Electrostatics interaction: Ewald
 765  * VdW interaction:            LJEwald
 766  * Geometry:                   Water3-Particle
 767  * Calculate force/pot:        Force
 768  */
 769 void
 770 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_sse4_1_single
 771                     (t_nblist                    * gmx_restrict       nlist,
 772                      rvec                        * gmx_restrict          xx,
 773                      rvec                        * gmx_restrict          ff,
 774                      t_forcerec                  * gmx_restrict          fr,
 775                      t_mdatoms                   * gmx_restrict     mdatoms,
 776                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 777                      t_nrnb                      * gmx_restrict        nrnb)
 778 {
 779     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 780      * just 0 for non-waters.
 781      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 782      * jnr indices corresponding to data put in the four positions in the SIMD register.
 783      */
 784     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 785     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 786     int              jnrA,jnrB,jnrC,jnrD;
 787     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 788     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 789     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 790     real             rcutoff_scalar;
 791     real             *shiftvec,*fshift,*x,*f;
 792     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 793     real             scratch[4*DIM];
 794     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 795     int              vdwioffset0;
 796     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 797     int              vdwioffset1;
 798     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 799     int              vdwioffset2;
 800     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 801     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 802     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 803     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 804     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 805     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 806     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 807     real             *charge;
 808     int              nvdwtype;
 809     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 810     int              *vdwtype;
 811     real             *vdwparam;
 812     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 813     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 814     __m128           c6grid_00;
 815     __m128           c6grid_10;
 816     __m128           c6grid_20;
 817     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 818     real             *vdwgridparam;
 819     __m128           one_half  = _mm_set1_ps(0.5);
 820     __m128           minus_one = _mm_set1_ps(-1.0);
 821     __m128i          ewitab;
 822     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 823     real             *ewtab;
 824     __m128           dummy_mask,cutoff_mask;
 825     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 826     __m128           one     = _mm_set1_ps(1.0);
 827     __m128           two     = _mm_set1_ps(2.0);
 828     x                = xx[0];
 829     f                = ff[0];
 830
 831     nri              = nlist->nri;
 832     iinr             = nlist->iinr;
 833     jindex           = nlist->jindex;
 834     jjnr             = nlist->jjnr;
 835     shiftidx         = nlist->shift;
 836     gid              = nlist->gid;
 837     shiftvec         = fr->shift_vec[0];
 838     fshift           = fr->fshift[0];
 839     facel            = _mm_set1_ps(fr->epsfac);
 840     charge           = mdatoms->chargeA;
 841     nvdwtype         = fr->ntype;
 842     vdwparam         = fr->nbfp;
 843     vdwtype          = mdatoms->typeA;
 844     vdwgridparam     = fr->ljpme_c6grid;
 845     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 846     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 847     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 848
 849     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 850     ewtab            = fr->ic->tabq_coul_F;
 851     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 852     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 853
 854     /* Setup water-specific parameters */
 855     inr              = nlist->iinr[0];
 856     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 857     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 858     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 859     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 860
 861     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 862     rcutoff_scalar   = fr->rcoulomb;
 863     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 864     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 865
 866     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 867     rvdw             = _mm_set1_ps(fr->rvdw);
 868
 869     /* Avoid stupid compiler warnings */
 870     jnrA = jnrB = jnrC = jnrD = 0;
 871     j_coord_offsetA = 0;
 872     j_coord_offsetB = 0;
 873     j_coord_offsetC = 0;
 874     j_coord_offsetD = 0;
 875
 876     outeriter        = 0;
 877     inneriter        = 0;
 878
 879     for(iidx=0;iidx<4*DIM;iidx++)
 880     {
 881         scratch[iidx] = 0.0;
 882     }
 883
 884     /* Start outer loop over neighborlists */
 885     for(iidx=0; iidx<nri; iidx++)
 886     {
 887         /* Load shift vector for this list */
 888         i_shift_offset   = DIM*shiftidx[iidx];
 889
 890         /* Load limits for loop over neighbors */
 891         j_index_start    = jindex[iidx];
 892         j_index_end      = jindex[iidx+1];
 893
 894         /* Get outer coordinate index */
 895         inr              = iinr[iidx];
 896         i_coord_offset   = DIM*inr;
 897
 898         /* Load i particle coords and add shift vector */
 899         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 900                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 901
 902         fix0             = _mm_setzero_ps();
 903         fiy0             = _mm_setzero_ps();
 904         fiz0             = _mm_setzero_ps();
 905         fix1             = _mm_setzero_ps();
 906         fiy1             = _mm_setzero_ps();
 907         fiz1             = _mm_setzero_ps();
 908         fix2             = _mm_setzero_ps();
 909         fiy2             = _mm_setzero_ps();
 910         fiz2             = _mm_setzero_ps();
 911
 912         /* Start inner kernel loop */
 913         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 914         {
 915
 916             /* Get j neighbor index, and coordinate index */
 917             jnrA             = jjnr[jidx];
 918             jnrB             = jjnr[jidx+1];
 919             jnrC             = jjnr[jidx+2];
 920             jnrD             = jjnr[jidx+3];
 921             j_coord_offsetA  = DIM*jnrA;
 922             j_coord_offsetB  = DIM*jnrB;
 923             j_coord_offsetC  = DIM*jnrC;
 924             j_coord_offsetD  = DIM*jnrD;
 925
 926             /* load j atom coordinates */
 927             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 928                                               x+j_coord_offsetC,x+j_coord_offsetD,
 929                                               &jx0,&jy0,&jz0);
 930
 931             /* Calculate displacement vector */
 932             dx00             = _mm_sub_ps(ix0,jx0);
 933             dy00             = _mm_sub_ps(iy0,jy0);
 934             dz00             = _mm_sub_ps(iz0,jz0);
 935             dx10             = _mm_sub_ps(ix1,jx0);
 936             dy10             = _mm_sub_ps(iy1,jy0);
 937             dz10             = _mm_sub_ps(iz1,jz0);
 938             dx20             = _mm_sub_ps(ix2,jx0);
 939             dy20             = _mm_sub_ps(iy2,jy0);
 940             dz20             = _mm_sub_ps(iz2,jz0);
 941
 942             /* Calculate squared distance and things based on it */
 943             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 944             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 945             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 946
 947             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 948             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 949             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 950
 951             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 952             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 953             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 954
 955             /* Load parameters for j particles */
 956             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 957                                                               charge+jnrC+0,charge+jnrD+0);
 958             vdwjidx0A        = 2*vdwtype[jnrA+0];
 959             vdwjidx0B        = 2*vdwtype[jnrB+0];
 960             vdwjidx0C        = 2*vdwtype[jnrC+0];
 961             vdwjidx0D        = 2*vdwtype[jnrD+0];
 962
 963             fjx0             = _mm_setzero_ps();
 964             fjy0             = _mm_setzero_ps();
 965             fjz0             = _mm_setzero_ps();
 966
 967             /**************************
 968              * CALCULATE INTERACTIONS *
 969              **************************/
 970
 971             if (gmx_mm_any_lt(rsq00,rcutoff2))
 972             {
 973
 974             r00              = _mm_mul_ps(rsq00,rinv00);
 975
 976             /* Compute parameters for interactions between i and j atoms */
 977             qq00             = _mm_mul_ps(iq0,jq0);
 978             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 979                                          vdwparam+vdwioffset0+vdwjidx0B,
 980                                          vdwparam+vdwioffset0+vdwjidx0C,
 981                                          vdwparam+vdwioffset0+vdwjidx0D,
 982                                          &c6_00,&c12_00);
 983
 984             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 985                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 986                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 987                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 988
 989             /* EWALD ELECTROSTATICS */
 990
 991             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 992             ewrt             = _mm_mul_ps(r00,ewtabscale);
 993             ewitab           = _mm_cvttps_epi32(ewrt);
 994             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 995             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 996                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 997                                          &ewtabF,&ewtabFn);
 998             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 999             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1000
1001             /* Analytical LJ-PME */
1002             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1003             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1004             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1005             exponent         = gmx_simd_exp_r(ewcljrsq);
1006             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1007             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1008             /* f6A = 6 * C6grid * (1 - poly) */
1009             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1010             /* f6B = C6grid * exponent * beta^6 */
1011             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1012             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1013             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);
1014
1015             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1016
1017             fscal            = _mm_add_ps(felec,fvdw);
1018
1019             fscal            = _mm_and_ps(fscal,cutoff_mask);
1020
1021             /* Calculate temporary vectorial force */
1022             tx               = _mm_mul_ps(fscal,dx00);
1023             ty               = _mm_mul_ps(fscal,dy00);
1024             tz               = _mm_mul_ps(fscal,dz00);
1025
1026             /* Update vectorial force */
1027             fix0             = _mm_add_ps(fix0,tx);
1028             fiy0             = _mm_add_ps(fiy0,ty);
1029             fiz0             = _mm_add_ps(fiz0,tz);
1030
1031             fjx0             = _mm_add_ps(fjx0,tx);
1032             fjy0             = _mm_add_ps(fjy0,ty);
1033             fjz0             = _mm_add_ps(fjz0,tz);
1034
1035             }
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             if (gmx_mm_any_lt(rsq10,rcutoff2))
1042             {
1043
1044             r10              = _mm_mul_ps(rsq10,rinv10);
1045
1046             /* Compute parameters for interactions between i and j atoms */
1047             qq10             = _mm_mul_ps(iq1,jq0);
1048
1049             /* EWALD ELECTROSTATICS */
1050
1051             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1052             ewrt             = _mm_mul_ps(r10,ewtabscale);
1053             ewitab           = _mm_cvttps_epi32(ewrt);
1054             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1055             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1056                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1057                                          &ewtabF,&ewtabFn);
1058             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1059             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1060
1061             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1062
1063             fscal            = felec;
1064
1065             fscal            = _mm_and_ps(fscal,cutoff_mask);
1066
1067             /* Calculate temporary vectorial force */
1068             tx               = _mm_mul_ps(fscal,dx10);
1069             ty               = _mm_mul_ps(fscal,dy10);
1070             tz               = _mm_mul_ps(fscal,dz10);
1071
1072             /* Update vectorial force */
1073             fix1             = _mm_add_ps(fix1,tx);
1074             fiy1             = _mm_add_ps(fiy1,ty);
1075             fiz1             = _mm_add_ps(fiz1,tz);
1076
1077             fjx0             = _mm_add_ps(fjx0,tx);
1078             fjy0             = _mm_add_ps(fjy0,ty);
1079             fjz0             = _mm_add_ps(fjz0,tz);
1080
1081             }
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             if (gmx_mm_any_lt(rsq20,rcutoff2))
1088             {
1089
1090             r20              = _mm_mul_ps(rsq20,rinv20);
1091
1092             /* Compute parameters for interactions between i and j atoms */
1093             qq20             = _mm_mul_ps(iq2,jq0);
1094
1095             /* EWALD ELECTROSTATICS */
1096
1097             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1098             ewrt             = _mm_mul_ps(r20,ewtabscale);
1099             ewitab           = _mm_cvttps_epi32(ewrt);
1100             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1101             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1102                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1103                                          &ewtabF,&ewtabFn);
1104             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1105             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1106
1107             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1108
1109             fscal            = felec;
1110
1111             fscal            = _mm_and_ps(fscal,cutoff_mask);
1112
1113             /* Calculate temporary vectorial force */
1114             tx               = _mm_mul_ps(fscal,dx20);
1115             ty               = _mm_mul_ps(fscal,dy20);
1116             tz               = _mm_mul_ps(fscal,dz20);
1117
1118             /* Update vectorial force */
1119             fix2             = _mm_add_ps(fix2,tx);
1120             fiy2             = _mm_add_ps(fiy2,ty);
1121             fiz2             = _mm_add_ps(fiz2,tz);
1122
1123             fjx0             = _mm_add_ps(fjx0,tx);
1124             fjy0             = _mm_add_ps(fjy0,ty);
1125             fjz0             = _mm_add_ps(fjz0,tz);
1126
1127             }
1128
1129             fjptrA             = f+j_coord_offsetA;
1130             fjptrB             = f+j_coord_offsetB;
1131             fjptrC             = f+j_coord_offsetC;
1132             fjptrD             = f+j_coord_offsetD;
1133
1134             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1135
1136             /* Inner loop uses 140 flops */
1137         }
1138
1139         if(jidx<j_index_end)
1140         {
1141
1142             /* Get j neighbor index, and coordinate index */
1143             jnrlistA         = jjnr[jidx];
1144             jnrlistB         = jjnr[jidx+1];
1145             jnrlistC         = jjnr[jidx+2];
1146             jnrlistD         = jjnr[jidx+3];
1147             /* Sign of each element will be negative for non-real atoms.
1148              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1149              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1150              */
1151             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1152             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1153             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1154             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1155             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1156             j_coord_offsetA  = DIM*jnrA;
1157             j_coord_offsetB  = DIM*jnrB;
1158             j_coord_offsetC  = DIM*jnrC;
1159             j_coord_offsetD  = DIM*jnrD;
1160
1161             /* load j atom coordinates */
1162             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1163                                               x+j_coord_offsetC,x+j_coord_offsetD,
1164                                               &jx0,&jy0,&jz0);
1165
1166             /* Calculate displacement vector */
1167             dx00             = _mm_sub_ps(ix0,jx0);
1168             dy00             = _mm_sub_ps(iy0,jy0);
1169             dz00             = _mm_sub_ps(iz0,jz0);
1170             dx10             = _mm_sub_ps(ix1,jx0);
1171             dy10             = _mm_sub_ps(iy1,jy0);
1172             dz10             = _mm_sub_ps(iz1,jz0);
1173             dx20             = _mm_sub_ps(ix2,jx0);
1174             dy20             = _mm_sub_ps(iy2,jy0);
1175             dz20             = _mm_sub_ps(iz2,jz0);
1176
1177             /* Calculate squared distance and things based on it */
1178             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1179             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1180             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1181
1182             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1183             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1184             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1185
1186             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1187             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1188             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1189
1190             /* Load parameters for j particles */
1191             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1192                                                               charge+jnrC+0,charge+jnrD+0);
1193             vdwjidx0A        = 2*vdwtype[jnrA+0];
1194             vdwjidx0B        = 2*vdwtype[jnrB+0];
1195             vdwjidx0C        = 2*vdwtype[jnrC+0];
1196             vdwjidx0D        = 2*vdwtype[jnrD+0];
1197
1198             fjx0             = _mm_setzero_ps();
1199             fjy0             = _mm_setzero_ps();
1200             fjz0             = _mm_setzero_ps();
1201
1202             /**************************
1203              * CALCULATE INTERACTIONS *
1204              **************************/
1205
1206             if (gmx_mm_any_lt(rsq00,rcutoff2))
1207             {
1208
1209             r00              = _mm_mul_ps(rsq00,rinv00);
1210             r00              = _mm_andnot_ps(dummy_mask,r00);
1211
1212             /* Compute parameters for interactions between i and j atoms */
1213             qq00             = _mm_mul_ps(iq0,jq0);
1214             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1215                                          vdwparam+vdwioffset0+vdwjidx0B,
1216                                          vdwparam+vdwioffset0+vdwjidx0C,
1217                                          vdwparam+vdwioffset0+vdwjidx0D,
1218                                          &c6_00,&c12_00);
1219
1220             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1221                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1222                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1223                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1224
1225             /* EWALD ELECTROSTATICS */
1226
1227             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1228             ewrt             = _mm_mul_ps(r00,ewtabscale);
1229             ewitab           = _mm_cvttps_epi32(ewrt);
1230             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1231             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1232                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1233                                          &ewtabF,&ewtabFn);
1234             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1235             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1236
1237             /* Analytical LJ-PME */
1238             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1239             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1240             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1241             exponent         = gmx_simd_exp_r(ewcljrsq);
1242             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1243             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1244             /* f6A = 6 * C6grid * (1 - poly) */
1245             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1246             /* f6B = C6grid * exponent * beta^6 */
1247             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1248             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1249             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);
1250
1251             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1252
1253             fscal            = _mm_add_ps(felec,fvdw);
1254
1255             fscal            = _mm_and_ps(fscal,cutoff_mask);
1256
1257             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1258
1259             /* Calculate temporary vectorial force */
1260             tx               = _mm_mul_ps(fscal,dx00);
1261             ty               = _mm_mul_ps(fscal,dy00);
1262             tz               = _mm_mul_ps(fscal,dz00);
1263
1264             /* Update vectorial force */
1265             fix0             = _mm_add_ps(fix0,tx);
1266             fiy0             = _mm_add_ps(fiy0,ty);
1267             fiz0             = _mm_add_ps(fiz0,tz);
1268
1269             fjx0             = _mm_add_ps(fjx0,tx);
1270             fjy0             = _mm_add_ps(fjy0,ty);
1271             fjz0             = _mm_add_ps(fjz0,tz);
1272
1273             }
1274
1275             /**************************
1276              * CALCULATE INTERACTIONS *
1277              **************************/
1278
1279             if (gmx_mm_any_lt(rsq10,rcutoff2))
1280             {
1281
1282             r10              = _mm_mul_ps(rsq10,rinv10);
1283             r10              = _mm_andnot_ps(dummy_mask,r10);
1284
1285             /* Compute parameters for interactions between i and j atoms */
1286             qq10             = _mm_mul_ps(iq1,jq0);
1287
1288             /* EWALD ELECTROSTATICS */
1289
1290             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1291             ewrt             = _mm_mul_ps(r10,ewtabscale);
1292             ewitab           = _mm_cvttps_epi32(ewrt);
1293             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1294             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1295                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1296                                          &ewtabF,&ewtabFn);
1297             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1298             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1299
1300             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1301
1302             fscal            = felec;
1303
1304             fscal            = _mm_and_ps(fscal,cutoff_mask);
1305
1306             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1307
1308             /* Calculate temporary vectorial force */
1309             tx               = _mm_mul_ps(fscal,dx10);
1310             ty               = _mm_mul_ps(fscal,dy10);
1311             tz               = _mm_mul_ps(fscal,dz10);
1312
1313             /* Update vectorial force */
1314             fix1             = _mm_add_ps(fix1,tx);
1315             fiy1             = _mm_add_ps(fiy1,ty);
1316             fiz1             = _mm_add_ps(fiz1,tz);
1317
1318             fjx0             = _mm_add_ps(fjx0,tx);
1319             fjy0             = _mm_add_ps(fjy0,ty);
1320             fjz0             = _mm_add_ps(fjz0,tz);
1321
1322             }
1323
1324             /**************************
1325              * CALCULATE INTERACTIONS *
1326              **************************/
1327
1328             if (gmx_mm_any_lt(rsq20,rcutoff2))
1329             {
1330
1331             r20              = _mm_mul_ps(rsq20,rinv20);
1332             r20              = _mm_andnot_ps(dummy_mask,r20);
1333
1334             /* Compute parameters for interactions between i and j atoms */
1335             qq20             = _mm_mul_ps(iq2,jq0);
1336
1337             /* EWALD ELECTROSTATICS */
1338
1339             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1340             ewrt             = _mm_mul_ps(r20,ewtabscale);
1341             ewitab           = _mm_cvttps_epi32(ewrt);
1342             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1343             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1344                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1345                                          &ewtabF,&ewtabFn);
1346             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1347             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1348
1349             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1350
1351             fscal            = felec;
1352
1353             fscal            = _mm_and_ps(fscal,cutoff_mask);
1354
1355             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1356
1357             /* Calculate temporary vectorial force */
1358             tx               = _mm_mul_ps(fscal,dx20);
1359             ty               = _mm_mul_ps(fscal,dy20);
1360             tz               = _mm_mul_ps(fscal,dz20);
1361
1362             /* Update vectorial force */
1363             fix2             = _mm_add_ps(fix2,tx);
1364             fiy2             = _mm_add_ps(fiy2,ty);
1365             fiz2             = _mm_add_ps(fiz2,tz);
1366
1367             fjx0             = _mm_add_ps(fjx0,tx);
1368             fjy0             = _mm_add_ps(fjy0,ty);
1369             fjz0             = _mm_add_ps(fjz0,tz);
1370
1371             }
1372
1373             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1374             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1375             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1376             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1377
1378             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1379
1380             /* Inner loop uses 143 flops */
1381         }
1382
1383         /* End of innermost loop */
1384
1385         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1386                                               f+i_coord_offset,fshift+i_shift_offset);
1387
1388         /* Increment number of inner iterations */
1389         inneriter                  += j_index_end - j_index_start;
1390
1391         /* Outer loop uses 18 flops */
1392     }
1393
1394     /* Increment number of outer iterations */
1395     outeriter        += nri;
1396
1397     /* Update outer/inner flops */
1398
1399     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*143);
1400 }