src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128           c6grid_00;
 106     __m128           c6grid_10;
 107     __m128           c6grid_20;
 108     __m128           c6grid_30;
 109     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 110     real             *vdwgridparam;
 111     __m128           one_half = _mm_set1_ps(0.5);
 112     __m128           minus_one = _mm_set1_ps(-1.0);
 113     __m128i          ewitab;
 114     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 115     real             *ewtab;
 116     __m128           dummy_mask,cutoff_mask;
 117     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 118     __m128           one     = _mm_set1_ps(1.0);
 119     __m128           two     = _mm_set1_ps(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm_set1_ps(fr->epsfac);
 132     charge           = mdatoms->chargeA;
 133     nvdwtype         = fr->ntype;
 134     vdwparam         = fr->nbfp;
 135     vdwtype          = mdatoms->typeA;
 136     vdwgridparam     = fr->ljpme_c6grid;
 137     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 138     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 139     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 140
 141     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 142     ewtab            = fr->ic->tabq_coul_FDV0;
 143     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 144     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 145
 146     /* Setup water-specific parameters */
 147     inr              = nlist->iinr[0];
 148     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 149     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 150     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 151     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 154     rcutoff_scalar   = fr->rcoulomb;
 155     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 156     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 157
 158     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 159     rvdw             = _mm_set1_ps(fr->rvdw);
 160
 161     /* Avoid stupid compiler warnings */
 162     jnrA = jnrB = jnrC = jnrD = 0;
 163     j_coord_offsetA = 0;
 164     j_coord_offsetB = 0;
 165     j_coord_offsetC = 0;
 166     j_coord_offsetD = 0;
 167
 168     outeriter        = 0;
 169     inneriter        = 0;
 170
 171     for(iidx=0;iidx<4*DIM;iidx++)
 172     {
 173         scratch[iidx] = 0.0;
 174     }
 175
 176     /* Start outer loop over neighborlists */
 177     for(iidx=0; iidx<nri; iidx++)
 178     {
 179         /* Load shift vector for this list */
 180         i_shift_offset   = DIM*shiftidx[iidx];
 181
 182         /* Load limits for loop over neighbors */
 183         j_index_start    = jindex[iidx];
 184         j_index_end      = jindex[iidx+1];
 185
 186         /* Get outer coordinate index */
 187         inr              = iinr[iidx];
 188         i_coord_offset   = DIM*inr;
 189
 190         /* Load i particle coords and add shift vector */
 191         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 192                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 193
 194         fix0             = _mm_setzero_ps();
 195         fiy0             = _mm_setzero_ps();
 196         fiz0             = _mm_setzero_ps();
 197         fix1             = _mm_setzero_ps();
 198         fiy1             = _mm_setzero_ps();
 199         fiz1             = _mm_setzero_ps();
 200         fix2             = _mm_setzero_ps();
 201         fiy2             = _mm_setzero_ps();
 202         fiz2             = _mm_setzero_ps();
 203         fix3             = _mm_setzero_ps();
 204         fiy3             = _mm_setzero_ps();
 205         fiz3             = _mm_setzero_ps();
 206
 207         /* Reset potential sums */
 208         velecsum         = _mm_setzero_ps();
 209         vvdwsum          = _mm_setzero_ps();
 210
 211         /* Start inner kernel loop */
 212         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 213         {
 214
 215             /* Get j neighbor index, and coordinate index */
 216             jnrA             = jjnr[jidx];
 217             jnrB             = jjnr[jidx+1];
 218             jnrC             = jjnr[jidx+2];
 219             jnrD             = jjnr[jidx+3];
 220             j_coord_offsetA  = DIM*jnrA;
 221             j_coord_offsetB  = DIM*jnrB;
 222             j_coord_offsetC  = DIM*jnrC;
 223             j_coord_offsetD  = DIM*jnrD;
 224
 225             /* load j atom coordinates */
 226             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 227                                               x+j_coord_offsetC,x+j_coord_offsetD,
 228                                               &jx0,&jy0,&jz0);
 229
 230             /* Calculate displacement vector */
 231             dx00             = _mm_sub_ps(ix0,jx0);
 232             dy00             = _mm_sub_ps(iy0,jy0);
 233             dz00             = _mm_sub_ps(iz0,jz0);
 234             dx10             = _mm_sub_ps(ix1,jx0);
 235             dy10             = _mm_sub_ps(iy1,jy0);
 236             dz10             = _mm_sub_ps(iz1,jz0);
 237             dx20             = _mm_sub_ps(ix2,jx0);
 238             dy20             = _mm_sub_ps(iy2,jy0);
 239             dz20             = _mm_sub_ps(iz2,jz0);
 240             dx30             = _mm_sub_ps(ix3,jx0);
 241             dy30             = _mm_sub_ps(iy3,jy0);
 242             dz30             = _mm_sub_ps(iz3,jz0);
 243
 244             /* Calculate squared distance and things based on it */
 245             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 246             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 247             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 248             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 249
 250             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 251             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 252             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 253             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 254
 255             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 256             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 257             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 258             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 259
 260             /* Load parameters for j particles */
 261             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 262                                                               charge+jnrC+0,charge+jnrD+0);
 263             vdwjidx0A        = 2*vdwtype[jnrA+0];
 264             vdwjidx0B        = 2*vdwtype[jnrB+0];
 265             vdwjidx0C        = 2*vdwtype[jnrC+0];
 266             vdwjidx0D        = 2*vdwtype[jnrD+0];
 267
 268             fjx0             = _mm_setzero_ps();
 269             fjy0             = _mm_setzero_ps();
 270             fjz0             = _mm_setzero_ps();
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             if (gmx_mm_any_lt(rsq00,rcutoff2))
 277             {
 278
 279             r00              = _mm_mul_ps(rsq00,rinv00);
 280
 281             /* Compute parameters for interactions between i and j atoms */
 282             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 283                                          vdwparam+vdwioffset0+vdwjidx0B,
 284                                          vdwparam+vdwioffset0+vdwjidx0C,
 285                                          vdwparam+vdwioffset0+vdwjidx0D,
 286                                          &c6_00,&c12_00);
 287             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 288                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 289                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 290                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 291
 292             /* Analytical LJ-PME */
 293             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 294             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 295             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 296             exponent         = gmx_simd_exp_r(ewcljrsq);
 297             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 298             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 299             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 300             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 301             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 302             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) ,
 303                                           _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));
 304             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 305             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);
 306
 307             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 311             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 312
 313             fscal            = fvdw;
 314
 315             fscal            = _mm_and_ps(fscal,cutoff_mask);
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm_mul_ps(fscal,dx00);
 319             ty               = _mm_mul_ps(fscal,dy00);
 320             tz               = _mm_mul_ps(fscal,dz00);
 321
 322             /* Update vectorial force */
 323             fix0             = _mm_add_ps(fix0,tx);
 324             fiy0             = _mm_add_ps(fiy0,ty);
 325             fiz0             = _mm_add_ps(fiz0,tz);
 326
 327             fjx0             = _mm_add_ps(fjx0,tx);
 328             fjy0             = _mm_add_ps(fjy0,ty);
 329             fjz0             = _mm_add_ps(fjz0,tz);
 330
 331             }
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             if (gmx_mm_any_lt(rsq10,rcutoff2))
 338             {
 339
 340             r10              = _mm_mul_ps(rsq10,rinv10);
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq10             = _mm_mul_ps(iq1,jq0);
 344
 345             /* EWALD ELECTROSTATICS */
 346
 347             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 348             ewrt             = _mm_mul_ps(r10,ewtabscale);
 349             ewitab           = _mm_cvttps_epi32(ewrt);
 350             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 351             ewitab           = _mm_slli_epi32(ewitab,2);
 352             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 353             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 354             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 355             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 356             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 357             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 358             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 359             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 360             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 361
 362             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velec            = _mm_and_ps(velec,cutoff_mask);
 366             velecsum         = _mm_add_ps(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             fscal            = _mm_and_ps(fscal,cutoff_mask);
 371
 372             /* Calculate temporary vectorial force */
 373             tx               = _mm_mul_ps(fscal,dx10);
 374             ty               = _mm_mul_ps(fscal,dy10);
 375             tz               = _mm_mul_ps(fscal,dz10);
 376
 377             /* Update vectorial force */
 378             fix1             = _mm_add_ps(fix1,tx);
 379             fiy1             = _mm_add_ps(fiy1,ty);
 380             fiz1             = _mm_add_ps(fiz1,tz);
 381
 382             fjx0             = _mm_add_ps(fjx0,tx);
 383             fjy0             = _mm_add_ps(fjy0,ty);
 384             fjz0             = _mm_add_ps(fjz0,tz);
 385
 386             }
 387
 388             /**************************
 389              * CALCULATE INTERACTIONS *
 390              **************************/
 391
 392             if (gmx_mm_any_lt(rsq20,rcutoff2))
 393             {
 394
 395             r20              = _mm_mul_ps(rsq20,rinv20);
 396
 397             /* Compute parameters for interactions between i and j atoms */
 398             qq20             = _mm_mul_ps(iq2,jq0);
 399
 400             /* EWALD ELECTROSTATICS */
 401
 402             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 403             ewrt             = _mm_mul_ps(r20,ewtabscale);
 404             ewitab           = _mm_cvttps_epi32(ewrt);
 405             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 406             ewitab           = _mm_slli_epi32(ewitab,2);
 407             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 408             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 409             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 410             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 411             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 412             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 413             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 414             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 415             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 416
 417             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velec            = _mm_and_ps(velec,cutoff_mask);
 421             velecsum         = _mm_add_ps(velecsum,velec);
 422
 423             fscal            = felec;
 424
 425             fscal            = _mm_and_ps(fscal,cutoff_mask);
 426
 427             /* Calculate temporary vectorial force */
 428             tx               = _mm_mul_ps(fscal,dx20);
 429             ty               = _mm_mul_ps(fscal,dy20);
 430             tz               = _mm_mul_ps(fscal,dz20);
 431
 432             /* Update vectorial force */
 433             fix2             = _mm_add_ps(fix2,tx);
 434             fiy2             = _mm_add_ps(fiy2,ty);
 435             fiz2             = _mm_add_ps(fiz2,tz);
 436
 437             fjx0             = _mm_add_ps(fjx0,tx);
 438             fjy0             = _mm_add_ps(fjy0,ty);
 439             fjz0             = _mm_add_ps(fjz0,tz);
 440
 441             }
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             if (gmx_mm_any_lt(rsq30,rcutoff2))
 448             {
 449
 450             r30              = _mm_mul_ps(rsq30,rinv30);
 451
 452             /* Compute parameters for interactions between i and j atoms */
 453             qq30             = _mm_mul_ps(iq3,jq0);
 454
 455             /* EWALD ELECTROSTATICS */
 456
 457             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 458             ewrt             = _mm_mul_ps(r30,ewtabscale);
 459             ewitab           = _mm_cvttps_epi32(ewrt);
 460             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 461             ewitab           = _mm_slli_epi32(ewitab,2);
 462             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 463             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 464             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 465             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 466             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 467             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 468             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 469             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 470             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 471
 472             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 473
 474             /* Update potential sum for this i atom from the interaction with this j atom. */
 475             velec            = _mm_and_ps(velec,cutoff_mask);
 476             velecsum         = _mm_add_ps(velecsum,velec);
 477
 478             fscal            = felec;
 479
 480             fscal            = _mm_and_ps(fscal,cutoff_mask);
 481
 482             /* Calculate temporary vectorial force */
 483             tx               = _mm_mul_ps(fscal,dx30);
 484             ty               = _mm_mul_ps(fscal,dy30);
 485             tz               = _mm_mul_ps(fscal,dz30);
 486
 487             /* Update vectorial force */
 488             fix3             = _mm_add_ps(fix3,tx);
 489             fiy3             = _mm_add_ps(fiy3,ty);
 490             fiz3             = _mm_add_ps(fiz3,tz);
 491
 492             fjx0             = _mm_add_ps(fjx0,tx);
 493             fjy0             = _mm_add_ps(fjy0,ty);
 494             fjz0             = _mm_add_ps(fjz0,tz);
 495
 496             }
 497
 498             fjptrA             = f+j_coord_offsetA;
 499             fjptrB             = f+j_coord_offsetB;
 500             fjptrC             = f+j_coord_offsetC;
 501             fjptrD             = f+j_coord_offsetD;
 502
 503             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 504
 505             /* Inner loop uses 200 flops */
 506         }
 507
 508         if(jidx<j_index_end)
 509         {
 510
 511             /* Get j neighbor index, and coordinate index */
 512             jnrlistA         = jjnr[jidx];
 513             jnrlistB         = jjnr[jidx+1];
 514             jnrlistC         = jjnr[jidx+2];
 515             jnrlistD         = jjnr[jidx+3];
 516             /* Sign of each element will be negative for non-real atoms.
 517              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 518              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 519              */
 520             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 521             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 522             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 523             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 524             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 525             j_coord_offsetA  = DIM*jnrA;
 526             j_coord_offsetB  = DIM*jnrB;
 527             j_coord_offsetC  = DIM*jnrC;
 528             j_coord_offsetD  = DIM*jnrD;
 529
 530             /* load j atom coordinates */
 531             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 532                                               x+j_coord_offsetC,x+j_coord_offsetD,
 533                                               &jx0,&jy0,&jz0);
 534
 535             /* Calculate displacement vector */
 536             dx00             = _mm_sub_ps(ix0,jx0);
 537             dy00             = _mm_sub_ps(iy0,jy0);
 538             dz00             = _mm_sub_ps(iz0,jz0);
 539             dx10             = _mm_sub_ps(ix1,jx0);
 540             dy10             = _mm_sub_ps(iy1,jy0);
 541             dz10             = _mm_sub_ps(iz1,jz0);
 542             dx20             = _mm_sub_ps(ix2,jx0);
 543             dy20             = _mm_sub_ps(iy2,jy0);
 544             dz20             = _mm_sub_ps(iz2,jz0);
 545             dx30             = _mm_sub_ps(ix3,jx0);
 546             dy30             = _mm_sub_ps(iy3,jy0);
 547             dz30             = _mm_sub_ps(iz3,jz0);
 548
 549             /* Calculate squared distance and things based on it */
 550             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 551             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 552             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 553             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 554
 555             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 556             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 557             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 558             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 559
 560             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 561             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 562             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 563             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 564
 565             /* Load parameters for j particles */
 566             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 567                                                               charge+jnrC+0,charge+jnrD+0);
 568             vdwjidx0A        = 2*vdwtype[jnrA+0];
 569             vdwjidx0B        = 2*vdwtype[jnrB+0];
 570             vdwjidx0C        = 2*vdwtype[jnrC+0];
 571             vdwjidx0D        = 2*vdwtype[jnrD+0];
 572
 573             fjx0             = _mm_setzero_ps();
 574             fjy0             = _mm_setzero_ps();
 575             fjz0             = _mm_setzero_ps();
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             if (gmx_mm_any_lt(rsq00,rcutoff2))
 582             {
 583
 584             r00              = _mm_mul_ps(rsq00,rinv00);
 585             r00              = _mm_andnot_ps(dummy_mask,r00);
 586
 587             /* Compute parameters for interactions between i and j atoms */
 588             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 589                                          vdwparam+vdwioffset0+vdwjidx0B,
 590                                          vdwparam+vdwioffset0+vdwjidx0C,
 591                                          vdwparam+vdwioffset0+vdwjidx0D,
 592                                          &c6_00,&c12_00);
 593             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 594                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 595                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 596                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 597
 598             /* Analytical LJ-PME */
 599             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 600             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 601             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 602             exponent         = gmx_simd_exp_r(ewcljrsq);
 603             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 604             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 605             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 606             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 607             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 608             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) ,
 609                                           _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));
 610             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 611             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);
 612
 613             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 614
 615             /* Update potential sum for this i atom from the interaction with this j atom. */
 616             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 617             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 618             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 619
 620             fscal            = fvdw;
 621
 622             fscal            = _mm_and_ps(fscal,cutoff_mask);
 623
 624             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 625
 626             /* Calculate temporary vectorial force */
 627             tx               = _mm_mul_ps(fscal,dx00);
 628             ty               = _mm_mul_ps(fscal,dy00);
 629             tz               = _mm_mul_ps(fscal,dz00);
 630
 631             /* Update vectorial force */
 632             fix0             = _mm_add_ps(fix0,tx);
 633             fiy0             = _mm_add_ps(fiy0,ty);
 634             fiz0             = _mm_add_ps(fiz0,tz);
 635
 636             fjx0             = _mm_add_ps(fjx0,tx);
 637             fjy0             = _mm_add_ps(fjy0,ty);
 638             fjz0             = _mm_add_ps(fjz0,tz);
 639
 640             }
 641
 642             /**************************
 643              * CALCULATE INTERACTIONS *
 644              **************************/
 645
 646             if (gmx_mm_any_lt(rsq10,rcutoff2))
 647             {
 648
 649             r10              = _mm_mul_ps(rsq10,rinv10);
 650             r10              = _mm_andnot_ps(dummy_mask,r10);
 651
 652             /* Compute parameters for interactions between i and j atoms */
 653             qq10             = _mm_mul_ps(iq1,jq0);
 654
 655             /* EWALD ELECTROSTATICS */
 656
 657             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 658             ewrt             = _mm_mul_ps(r10,ewtabscale);
 659             ewitab           = _mm_cvttps_epi32(ewrt);
 660             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 661             ewitab           = _mm_slli_epi32(ewitab,2);
 662             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 663             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 664             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 665             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 666             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 667             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 668             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 669             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 670             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 671
 672             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 673
 674             /* Update potential sum for this i atom from the interaction with this j atom. */
 675             velec            = _mm_and_ps(velec,cutoff_mask);
 676             velec            = _mm_andnot_ps(dummy_mask,velec);
 677             velecsum         = _mm_add_ps(velecsum,velec);
 678
 679             fscal            = felec;
 680
 681             fscal            = _mm_and_ps(fscal,cutoff_mask);
 682
 683             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 684
 685             /* Calculate temporary vectorial force */
 686             tx               = _mm_mul_ps(fscal,dx10);
 687             ty               = _mm_mul_ps(fscal,dy10);
 688             tz               = _mm_mul_ps(fscal,dz10);
 689
 690             /* Update vectorial force */
 691             fix1             = _mm_add_ps(fix1,tx);
 692             fiy1             = _mm_add_ps(fiy1,ty);
 693             fiz1             = _mm_add_ps(fiz1,tz);
 694
 695             fjx0             = _mm_add_ps(fjx0,tx);
 696             fjy0             = _mm_add_ps(fjy0,ty);
 697             fjz0             = _mm_add_ps(fjz0,tz);
 698
 699             }
 700
 701             /**************************
 702              * CALCULATE INTERACTIONS *
 703              **************************/
 704
 705             if (gmx_mm_any_lt(rsq20,rcutoff2))
 706             {
 707
 708             r20              = _mm_mul_ps(rsq20,rinv20);
 709             r20              = _mm_andnot_ps(dummy_mask,r20);
 710
 711             /* Compute parameters for interactions between i and j atoms */
 712             qq20             = _mm_mul_ps(iq2,jq0);
 713
 714             /* EWALD ELECTROSTATICS */
 715
 716             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 717             ewrt             = _mm_mul_ps(r20,ewtabscale);
 718             ewitab           = _mm_cvttps_epi32(ewrt);
 719             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 720             ewitab           = _mm_slli_epi32(ewitab,2);
 721             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 722             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 723             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 724             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 725             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 726             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 727             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 728             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 729             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 730
 731             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 732
 733             /* Update potential sum for this i atom from the interaction with this j atom. */
 734             velec            = _mm_and_ps(velec,cutoff_mask);
 735             velec            = _mm_andnot_ps(dummy_mask,velec);
 736             velecsum         = _mm_add_ps(velecsum,velec);
 737
 738             fscal            = felec;
 739
 740             fscal            = _mm_and_ps(fscal,cutoff_mask);
 741
 742             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 743
 744             /* Calculate temporary vectorial force */
 745             tx               = _mm_mul_ps(fscal,dx20);
 746             ty               = _mm_mul_ps(fscal,dy20);
 747             tz               = _mm_mul_ps(fscal,dz20);
 748
 749             /* Update vectorial force */
 750             fix2             = _mm_add_ps(fix2,tx);
 751             fiy2             = _mm_add_ps(fiy2,ty);
 752             fiz2             = _mm_add_ps(fiz2,tz);
 753
 754             fjx0             = _mm_add_ps(fjx0,tx);
 755             fjy0             = _mm_add_ps(fjy0,ty);
 756             fjz0             = _mm_add_ps(fjz0,tz);
 757
 758             }
 759
 760             /**************************
 761              * CALCULATE INTERACTIONS *
 762              **************************/
 763
 764             if (gmx_mm_any_lt(rsq30,rcutoff2))
 765             {
 766
 767             r30              = _mm_mul_ps(rsq30,rinv30);
 768             r30              = _mm_andnot_ps(dummy_mask,r30);
 769
 770             /* Compute parameters for interactions between i and j atoms */
 771             qq30             = _mm_mul_ps(iq3,jq0);
 772
 773             /* EWALD ELECTROSTATICS */
 774
 775             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 776             ewrt             = _mm_mul_ps(r30,ewtabscale);
 777             ewitab           = _mm_cvttps_epi32(ewrt);
 778             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 779             ewitab           = _mm_slli_epi32(ewitab,2);
 780             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 781             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 782             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 783             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 784             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 785             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 786             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 787             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 788             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 789
 790             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 791
 792             /* Update potential sum for this i atom from the interaction with this j atom. */
 793             velec            = _mm_and_ps(velec,cutoff_mask);
 794             velec            = _mm_andnot_ps(dummy_mask,velec);
 795             velecsum         = _mm_add_ps(velecsum,velec);
 796
 797             fscal            = felec;
 798
 799             fscal            = _mm_and_ps(fscal,cutoff_mask);
 800
 801             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 802
 803             /* Calculate temporary vectorial force */
 804             tx               = _mm_mul_ps(fscal,dx30);
 805             ty               = _mm_mul_ps(fscal,dy30);
 806             tz               = _mm_mul_ps(fscal,dz30);
 807
 808             /* Update vectorial force */
 809             fix3             = _mm_add_ps(fix3,tx);
 810             fiy3             = _mm_add_ps(fiy3,ty);
 811             fiz3             = _mm_add_ps(fiz3,tz);
 812
 813             fjx0             = _mm_add_ps(fjx0,tx);
 814             fjy0             = _mm_add_ps(fjy0,ty);
 815             fjz0             = _mm_add_ps(fjz0,tz);
 816
 817             }
 818
 819             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 820             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 821             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 822             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 823
 824             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 825
 826             /* Inner loop uses 204 flops */
 827         }
 828
 829         /* End of innermost loop */
 830
 831         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 832                                               f+i_coord_offset,fshift+i_shift_offset);
 833
 834         ggid                        = gid[iidx];
 835         /* Update potential energies */
 836         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 837         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 838
 839         /* Increment number of inner iterations */
 840         inneriter                  += j_index_end - j_index_start;
 841
 842         /* Outer loop uses 26 flops */
 843     }
 844
 845     /* Increment number of outer iterations */
 846     outeriter        += nri;
 847
 848     /* Update outer/inner flops */
 849
 850     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*204);
 851 }
 852 /*
 853  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_single
 854  * Electrostatics interaction: Ewald
 855  * VdW interaction:            LJEwald
 856  * Geometry:                   Water4-Particle
 857  * Calculate force/pot:        Force
 858  */
 859 void
 860 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_single
 861                     (t_nblist                    * gmx_restrict       nlist,
 862                      rvec                        * gmx_restrict          xx,
 863                      rvec                        * gmx_restrict          ff,
 864                      t_forcerec                  * gmx_restrict          fr,
 865                      t_mdatoms                   * gmx_restrict     mdatoms,
 866                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 867                      t_nrnb                      * gmx_restrict        nrnb)
 868 {
 869     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 870      * just 0 for non-waters.
 871      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 872      * jnr indices corresponding to data put in the four positions in the SIMD register.
 873      */
 874     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 875     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 876     int              jnrA,jnrB,jnrC,jnrD;
 877     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 878     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 879     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 880     real             rcutoff_scalar;
 881     real             *shiftvec,*fshift,*x,*f;
 882     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 883     real             scratch[4*DIM];
 884     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 885     int              vdwioffset0;
 886     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 887     int              vdwioffset1;
 888     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 889     int              vdwioffset2;
 890     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 891     int              vdwioffset3;
 892     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 893     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 894     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 895     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 896     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 897     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 898     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 899     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 900     real             *charge;
 901     int              nvdwtype;
 902     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 903     int              *vdwtype;
 904     real             *vdwparam;
 905     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 906     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 907     __m128           c6grid_00;
 908     __m128           c6grid_10;
 909     __m128           c6grid_20;
 910     __m128           c6grid_30;
 911     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 912     real             *vdwgridparam;
 913     __m128           one_half = _mm_set1_ps(0.5);
 914     __m128           minus_one = _mm_set1_ps(-1.0);
 915     __m128i          ewitab;
 916     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 917     real             *ewtab;
 918     __m128           dummy_mask,cutoff_mask;
 919     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 920     __m128           one     = _mm_set1_ps(1.0);
 921     __m128           two     = _mm_set1_ps(2.0);
 922     x                = xx[0];
 923     f                = ff[0];
 924
 925     nri              = nlist->nri;
 926     iinr             = nlist->iinr;
 927     jindex           = nlist->jindex;
 928     jjnr             = nlist->jjnr;
 929     shiftidx         = nlist->shift;
 930     gid              = nlist->gid;
 931     shiftvec         = fr->shift_vec[0];
 932     fshift           = fr->fshift[0];
 933     facel            = _mm_set1_ps(fr->epsfac);
 934     charge           = mdatoms->chargeA;
 935     nvdwtype         = fr->ntype;
 936     vdwparam         = fr->nbfp;
 937     vdwtype          = mdatoms->typeA;
 938     vdwgridparam     = fr->ljpme_c6grid;
 939     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 940     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 941     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 942
 943     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 944     ewtab            = fr->ic->tabq_coul_F;
 945     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 946     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 947
 948     /* Setup water-specific parameters */
 949     inr              = nlist->iinr[0];
 950     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 951     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 952     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 953     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 954
 955     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 956     rcutoff_scalar   = fr->rcoulomb;
 957     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 958     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 959
 960     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 961     rvdw             = _mm_set1_ps(fr->rvdw);
 962
 963     /* Avoid stupid compiler warnings */
 964     jnrA = jnrB = jnrC = jnrD = 0;
 965     j_coord_offsetA = 0;
 966     j_coord_offsetB = 0;
 967     j_coord_offsetC = 0;
 968     j_coord_offsetD = 0;
 969
 970     outeriter        = 0;
 971     inneriter        = 0;
 972
 973     for(iidx=0;iidx<4*DIM;iidx++)
 974     {
 975         scratch[iidx] = 0.0;
 976     }
 977
 978     /* Start outer loop over neighborlists */
 979     for(iidx=0; iidx<nri; iidx++)
 980     {
 981         /* Load shift vector for this list */
 982         i_shift_offset   = DIM*shiftidx[iidx];
 983
 984         /* Load limits for loop over neighbors */
 985         j_index_start    = jindex[iidx];
 986         j_index_end      = jindex[iidx+1];
 987
 988         /* Get outer coordinate index */
 989         inr              = iinr[iidx];
 990         i_coord_offset   = DIM*inr;
 991
 992         /* Load i particle coords and add shift vector */
 993         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 994                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 995
 996         fix0             = _mm_setzero_ps();
 997         fiy0             = _mm_setzero_ps();
 998         fiz0             = _mm_setzero_ps();
 999         fix1             = _mm_setzero_ps();
1000         fiy1             = _mm_setzero_ps();
1001         fiz1             = _mm_setzero_ps();
1002         fix2             = _mm_setzero_ps();
1003         fiy2             = _mm_setzero_ps();
1004         fiz2             = _mm_setzero_ps();
1005         fix3             = _mm_setzero_ps();
1006         fiy3             = _mm_setzero_ps();
1007         fiz3             = _mm_setzero_ps();
1008
1009         /* Start inner kernel loop */
1010         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1011         {
1012
1013             /* Get j neighbor index, and coordinate index */
1014             jnrA             = jjnr[jidx];
1015             jnrB             = jjnr[jidx+1];
1016             jnrC             = jjnr[jidx+2];
1017             jnrD             = jjnr[jidx+3];
1018             j_coord_offsetA  = DIM*jnrA;
1019             j_coord_offsetB  = DIM*jnrB;
1020             j_coord_offsetC  = DIM*jnrC;
1021             j_coord_offsetD  = DIM*jnrD;
1022
1023             /* load j atom coordinates */
1024             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1025                                               x+j_coord_offsetC,x+j_coord_offsetD,
1026                                               &jx0,&jy0,&jz0);
1027
1028             /* Calculate displacement vector */
1029             dx00             = _mm_sub_ps(ix0,jx0);
1030             dy00             = _mm_sub_ps(iy0,jy0);
1031             dz00             = _mm_sub_ps(iz0,jz0);
1032             dx10             = _mm_sub_ps(ix1,jx0);
1033             dy10             = _mm_sub_ps(iy1,jy0);
1034             dz10             = _mm_sub_ps(iz1,jz0);
1035             dx20             = _mm_sub_ps(ix2,jx0);
1036             dy20             = _mm_sub_ps(iy2,jy0);
1037             dz20             = _mm_sub_ps(iz2,jz0);
1038             dx30             = _mm_sub_ps(ix3,jx0);
1039             dy30             = _mm_sub_ps(iy3,jy0);
1040             dz30             = _mm_sub_ps(iz3,jz0);
1041
1042             /* Calculate squared distance and things based on it */
1043             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1044             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1045             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1046             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1047
1048             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1049             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1050             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1051             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1052
1053             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1054             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1055             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1056             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1057
1058             /* Load parameters for j particles */
1059             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1060                                                               charge+jnrC+0,charge+jnrD+0);
1061             vdwjidx0A        = 2*vdwtype[jnrA+0];
1062             vdwjidx0B        = 2*vdwtype[jnrB+0];
1063             vdwjidx0C        = 2*vdwtype[jnrC+0];
1064             vdwjidx0D        = 2*vdwtype[jnrD+0];
1065
1066             fjx0             = _mm_setzero_ps();
1067             fjy0             = _mm_setzero_ps();
1068             fjz0             = _mm_setzero_ps();
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             if (gmx_mm_any_lt(rsq00,rcutoff2))
1075             {
1076
1077             r00              = _mm_mul_ps(rsq00,rinv00);
1078
1079             /* Compute parameters for interactions between i and j atoms */
1080             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1081                                          vdwparam+vdwioffset0+vdwjidx0B,
1082                                          vdwparam+vdwioffset0+vdwjidx0C,
1083                                          vdwparam+vdwioffset0+vdwjidx0D,
1084                                          &c6_00,&c12_00);
1085             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1086                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1087                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1088                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1089
1090             /* Analytical LJ-PME */
1091             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1092             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1093             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1094             exponent         = gmx_simd_exp_r(ewcljrsq);
1095             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1096             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1097             /* f6A = 6 * C6grid * (1 - poly) */
1098             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1099             /* f6B = C6grid * exponent * beta^6 */
1100             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1101             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1102             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);
1103
1104             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1105
1106             fscal            = fvdw;
1107
1108             fscal            = _mm_and_ps(fscal,cutoff_mask);
1109
1110             /* Calculate temporary vectorial force */
1111             tx               = _mm_mul_ps(fscal,dx00);
1112             ty               = _mm_mul_ps(fscal,dy00);
1113             tz               = _mm_mul_ps(fscal,dz00);
1114
1115             /* Update vectorial force */
1116             fix0             = _mm_add_ps(fix0,tx);
1117             fiy0             = _mm_add_ps(fiy0,ty);
1118             fiz0             = _mm_add_ps(fiz0,tz);
1119
1120             fjx0             = _mm_add_ps(fjx0,tx);
1121             fjy0             = _mm_add_ps(fjy0,ty);
1122             fjz0             = _mm_add_ps(fjz0,tz);
1123
1124             }
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             if (gmx_mm_any_lt(rsq10,rcutoff2))
1131             {
1132
1133             r10              = _mm_mul_ps(rsq10,rinv10);
1134
1135             /* Compute parameters for interactions between i and j atoms */
1136             qq10             = _mm_mul_ps(iq1,jq0);
1137
1138             /* EWALD ELECTROSTATICS */
1139
1140             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1141             ewrt             = _mm_mul_ps(r10,ewtabscale);
1142             ewitab           = _mm_cvttps_epi32(ewrt);
1143             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1144             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1145                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1146                                          &ewtabF,&ewtabFn);
1147             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1148             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1149
1150             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1151
1152             fscal            = felec;
1153
1154             fscal            = _mm_and_ps(fscal,cutoff_mask);
1155
1156             /* Calculate temporary vectorial force */
1157             tx               = _mm_mul_ps(fscal,dx10);
1158             ty               = _mm_mul_ps(fscal,dy10);
1159             tz               = _mm_mul_ps(fscal,dz10);
1160
1161             /* Update vectorial force */
1162             fix1             = _mm_add_ps(fix1,tx);
1163             fiy1             = _mm_add_ps(fiy1,ty);
1164             fiz1             = _mm_add_ps(fiz1,tz);
1165
1166             fjx0             = _mm_add_ps(fjx0,tx);
1167             fjy0             = _mm_add_ps(fjy0,ty);
1168             fjz0             = _mm_add_ps(fjz0,tz);
1169
1170             }
1171
1172             /**************************
1173              * CALCULATE INTERACTIONS *
1174              **************************/
1175
1176             if (gmx_mm_any_lt(rsq20,rcutoff2))
1177             {
1178
1179             r20              = _mm_mul_ps(rsq20,rinv20);
1180
1181             /* Compute parameters for interactions between i and j atoms */
1182             qq20             = _mm_mul_ps(iq2,jq0);
1183
1184             /* EWALD ELECTROSTATICS */
1185
1186             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1187             ewrt             = _mm_mul_ps(r20,ewtabscale);
1188             ewitab           = _mm_cvttps_epi32(ewrt);
1189             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1190             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1191                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1192                                          &ewtabF,&ewtabFn);
1193             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1194             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1195
1196             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1197
1198             fscal            = felec;
1199
1200             fscal            = _mm_and_ps(fscal,cutoff_mask);
1201
1202             /* Calculate temporary vectorial force */
1203             tx               = _mm_mul_ps(fscal,dx20);
1204             ty               = _mm_mul_ps(fscal,dy20);
1205             tz               = _mm_mul_ps(fscal,dz20);
1206
1207             /* Update vectorial force */
1208             fix2             = _mm_add_ps(fix2,tx);
1209             fiy2             = _mm_add_ps(fiy2,ty);
1210             fiz2             = _mm_add_ps(fiz2,tz);
1211
1212             fjx0             = _mm_add_ps(fjx0,tx);
1213             fjy0             = _mm_add_ps(fjy0,ty);
1214             fjz0             = _mm_add_ps(fjz0,tz);
1215
1216             }
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             if (gmx_mm_any_lt(rsq30,rcutoff2))
1223             {
1224
1225             r30              = _mm_mul_ps(rsq30,rinv30);
1226
1227             /* Compute parameters for interactions between i and j atoms */
1228             qq30             = _mm_mul_ps(iq3,jq0);
1229
1230             /* EWALD ELECTROSTATICS */
1231
1232             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1233             ewrt             = _mm_mul_ps(r30,ewtabscale);
1234             ewitab           = _mm_cvttps_epi32(ewrt);
1235             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1236             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1237                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1238                                          &ewtabF,&ewtabFn);
1239             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1240             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1241
1242             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1243
1244             fscal            = felec;
1245
1246             fscal            = _mm_and_ps(fscal,cutoff_mask);
1247
1248             /* Calculate temporary vectorial force */
1249             tx               = _mm_mul_ps(fscal,dx30);
1250             ty               = _mm_mul_ps(fscal,dy30);
1251             tz               = _mm_mul_ps(fscal,dz30);
1252
1253             /* Update vectorial force */
1254             fix3             = _mm_add_ps(fix3,tx);
1255             fiy3             = _mm_add_ps(fiy3,ty);
1256             fiz3             = _mm_add_ps(fiz3,tz);
1257
1258             fjx0             = _mm_add_ps(fjx0,tx);
1259             fjy0             = _mm_add_ps(fjy0,ty);
1260             fjz0             = _mm_add_ps(fjz0,tz);
1261
1262             }
1263
1264             fjptrA             = f+j_coord_offsetA;
1265             fjptrB             = f+j_coord_offsetB;
1266             fjptrC             = f+j_coord_offsetC;
1267             fjptrD             = f+j_coord_offsetD;
1268
1269             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1270
1271             /* Inner loop uses 166 flops */
1272         }
1273
1274         if(jidx<j_index_end)
1275         {
1276
1277             /* Get j neighbor index, and coordinate index */
1278             jnrlistA         = jjnr[jidx];
1279             jnrlistB         = jjnr[jidx+1];
1280             jnrlistC         = jjnr[jidx+2];
1281             jnrlistD         = jjnr[jidx+3];
1282             /* Sign of each element will be negative for non-real atoms.
1283              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1284              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1285              */
1286             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1287             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1288             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1289             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1290             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1291             j_coord_offsetA  = DIM*jnrA;
1292             j_coord_offsetB  = DIM*jnrB;
1293             j_coord_offsetC  = DIM*jnrC;
1294             j_coord_offsetD  = DIM*jnrD;
1295
1296             /* load j atom coordinates */
1297             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1298                                               x+j_coord_offsetC,x+j_coord_offsetD,
1299                                               &jx0,&jy0,&jz0);
1300
1301             /* Calculate displacement vector */
1302             dx00             = _mm_sub_ps(ix0,jx0);
1303             dy00             = _mm_sub_ps(iy0,jy0);
1304             dz00             = _mm_sub_ps(iz0,jz0);
1305             dx10             = _mm_sub_ps(ix1,jx0);
1306             dy10             = _mm_sub_ps(iy1,jy0);
1307             dz10             = _mm_sub_ps(iz1,jz0);
1308             dx20             = _mm_sub_ps(ix2,jx0);
1309             dy20             = _mm_sub_ps(iy2,jy0);
1310             dz20             = _mm_sub_ps(iz2,jz0);
1311             dx30             = _mm_sub_ps(ix3,jx0);
1312             dy30             = _mm_sub_ps(iy3,jy0);
1313             dz30             = _mm_sub_ps(iz3,jz0);
1314
1315             /* Calculate squared distance and things based on it */
1316             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1317             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1318             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1319             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1320
1321             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1322             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1323             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1324             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1325
1326             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1327             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1328             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1329             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1330
1331             /* Load parameters for j particles */
1332             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1333                                                               charge+jnrC+0,charge+jnrD+0);
1334             vdwjidx0A        = 2*vdwtype[jnrA+0];
1335             vdwjidx0B        = 2*vdwtype[jnrB+0];
1336             vdwjidx0C        = 2*vdwtype[jnrC+0];
1337             vdwjidx0D        = 2*vdwtype[jnrD+0];
1338
1339             fjx0             = _mm_setzero_ps();
1340             fjy0             = _mm_setzero_ps();
1341             fjz0             = _mm_setzero_ps();
1342
1343             /**************************
1344              * CALCULATE INTERACTIONS *
1345              **************************/
1346
1347             if (gmx_mm_any_lt(rsq00,rcutoff2))
1348             {
1349
1350             r00              = _mm_mul_ps(rsq00,rinv00);
1351             r00              = _mm_andnot_ps(dummy_mask,r00);
1352
1353             /* Compute parameters for interactions between i and j atoms */
1354             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1355                                          vdwparam+vdwioffset0+vdwjidx0B,
1356                                          vdwparam+vdwioffset0+vdwjidx0C,
1357                                          vdwparam+vdwioffset0+vdwjidx0D,
1358                                          &c6_00,&c12_00);
1359             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1360                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1361                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1362                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1363
1364             /* Analytical LJ-PME */
1365             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1366             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1367             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1368             exponent         = gmx_simd_exp_r(ewcljrsq);
1369             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1370             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1371             /* f6A = 6 * C6grid * (1 - poly) */
1372             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1373             /* f6B = C6grid * exponent * beta^6 */
1374             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1375             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1376             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);
1377
1378             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1379
1380             fscal            = fvdw;
1381
1382             fscal            = _mm_and_ps(fscal,cutoff_mask);
1383
1384             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1385
1386             /* Calculate temporary vectorial force */
1387             tx               = _mm_mul_ps(fscal,dx00);
1388             ty               = _mm_mul_ps(fscal,dy00);
1389             tz               = _mm_mul_ps(fscal,dz00);
1390
1391             /* Update vectorial force */
1392             fix0             = _mm_add_ps(fix0,tx);
1393             fiy0             = _mm_add_ps(fiy0,ty);
1394             fiz0             = _mm_add_ps(fiz0,tz);
1395
1396             fjx0             = _mm_add_ps(fjx0,tx);
1397             fjy0             = _mm_add_ps(fjy0,ty);
1398             fjz0             = _mm_add_ps(fjz0,tz);
1399
1400             }
1401
1402             /**************************
1403              * CALCULATE INTERACTIONS *
1404              **************************/
1405
1406             if (gmx_mm_any_lt(rsq10,rcutoff2))
1407             {
1408
1409             r10              = _mm_mul_ps(rsq10,rinv10);
1410             r10              = _mm_andnot_ps(dummy_mask,r10);
1411
1412             /* Compute parameters for interactions between i and j atoms */
1413             qq10             = _mm_mul_ps(iq1,jq0);
1414
1415             /* EWALD ELECTROSTATICS */
1416
1417             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1418             ewrt             = _mm_mul_ps(r10,ewtabscale);
1419             ewitab           = _mm_cvttps_epi32(ewrt);
1420             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1421             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1422                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1423                                          &ewtabF,&ewtabFn);
1424             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1425             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1426
1427             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1428
1429             fscal            = felec;
1430
1431             fscal            = _mm_and_ps(fscal,cutoff_mask);
1432
1433             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1434
1435             /* Calculate temporary vectorial force */
1436             tx               = _mm_mul_ps(fscal,dx10);
1437             ty               = _mm_mul_ps(fscal,dy10);
1438             tz               = _mm_mul_ps(fscal,dz10);
1439
1440             /* Update vectorial force */
1441             fix1             = _mm_add_ps(fix1,tx);
1442             fiy1             = _mm_add_ps(fiy1,ty);
1443             fiz1             = _mm_add_ps(fiz1,tz);
1444
1445             fjx0             = _mm_add_ps(fjx0,tx);
1446             fjy0             = _mm_add_ps(fjy0,ty);
1447             fjz0             = _mm_add_ps(fjz0,tz);
1448
1449             }
1450
1451             /**************************
1452              * CALCULATE INTERACTIONS *
1453              **************************/
1454
1455             if (gmx_mm_any_lt(rsq20,rcutoff2))
1456             {
1457
1458             r20              = _mm_mul_ps(rsq20,rinv20);
1459             r20              = _mm_andnot_ps(dummy_mask,r20);
1460
1461             /* Compute parameters for interactions between i and j atoms */
1462             qq20             = _mm_mul_ps(iq2,jq0);
1463
1464             /* EWALD ELECTROSTATICS */
1465
1466             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1467             ewrt             = _mm_mul_ps(r20,ewtabscale);
1468             ewitab           = _mm_cvttps_epi32(ewrt);
1469             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1470             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1471                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1472                                          &ewtabF,&ewtabFn);
1473             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1474             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1475
1476             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1477
1478             fscal            = felec;
1479
1480             fscal            = _mm_and_ps(fscal,cutoff_mask);
1481
1482             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1483
1484             /* Calculate temporary vectorial force */
1485             tx               = _mm_mul_ps(fscal,dx20);
1486             ty               = _mm_mul_ps(fscal,dy20);
1487             tz               = _mm_mul_ps(fscal,dz20);
1488
1489             /* Update vectorial force */
1490             fix2             = _mm_add_ps(fix2,tx);
1491             fiy2             = _mm_add_ps(fiy2,ty);
1492             fiz2             = _mm_add_ps(fiz2,tz);
1493
1494             fjx0             = _mm_add_ps(fjx0,tx);
1495             fjy0             = _mm_add_ps(fjy0,ty);
1496             fjz0             = _mm_add_ps(fjz0,tz);
1497
1498             }
1499
1500             /**************************
1501              * CALCULATE INTERACTIONS *
1502              **************************/
1503
1504             if (gmx_mm_any_lt(rsq30,rcutoff2))
1505             {
1506
1507             r30              = _mm_mul_ps(rsq30,rinv30);
1508             r30              = _mm_andnot_ps(dummy_mask,r30);
1509
1510             /* Compute parameters for interactions between i and j atoms */
1511             qq30             = _mm_mul_ps(iq3,jq0);
1512
1513             /* EWALD ELECTROSTATICS */
1514
1515             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1516             ewrt             = _mm_mul_ps(r30,ewtabscale);
1517             ewitab           = _mm_cvttps_epi32(ewrt);
1518             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1519             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1520                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1521                                          &ewtabF,&ewtabFn);
1522             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1523             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1524
1525             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1526
1527             fscal            = felec;
1528
1529             fscal            = _mm_and_ps(fscal,cutoff_mask);
1530
1531             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1532
1533             /* Calculate temporary vectorial force */
1534             tx               = _mm_mul_ps(fscal,dx30);
1535             ty               = _mm_mul_ps(fscal,dy30);
1536             tz               = _mm_mul_ps(fscal,dz30);
1537
1538             /* Update vectorial force */
1539             fix3             = _mm_add_ps(fix3,tx);
1540             fiy3             = _mm_add_ps(fiy3,ty);
1541             fiz3             = _mm_add_ps(fiz3,tz);
1542
1543             fjx0             = _mm_add_ps(fjx0,tx);
1544             fjy0             = _mm_add_ps(fjy0,ty);
1545             fjz0             = _mm_add_ps(fjz0,tz);
1546
1547             }
1548
1549             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1550             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1551             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1552             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1553
1554             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1555
1556             /* Inner loop uses 170 flops */
1557         }
1558
1559         /* End of innermost loop */
1560
1561         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1562                                               f+i_coord_offset,fshift+i_shift_offset);
1563
1564         /* Increment number of inner iterations */
1565         inneriter                  += j_index_end - j_index_start;
1566
1567         /* Outer loop uses 24 flops */
1568     }
1569
1570     /* Increment number of outer iterations */
1571     outeriter        += nri;
1572
1573     /* Update outer/inner flops */
1574
1575     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*170);
1576 }