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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse2_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128           c6grid_00;
 108     __m128           c6grid_10;
 109     __m128           c6grid_20;
 110     __m128           c6grid_30;
 111     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 112     real             *vdwgridparam;
 113     __m128           one_half = _mm_set1_ps(0.5);
 114     __m128           minus_one = _mm_set1_ps(-1.0);
 115     __m128i          ewitab;
 116     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 117     real             *ewtab;
 118     __m128           dummy_mask,cutoff_mask;
 119     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 120     __m128           one     = _mm_set1_ps(1.0);
 121     __m128           two     = _mm_set1_ps(2.0);
 122     x                = xx[0];
 123     f                = ff[0];
 124
 125     nri              = nlist->nri;
 126     iinr             = nlist->iinr;
 127     jindex           = nlist->jindex;
 128     jjnr             = nlist->jjnr;
 129     shiftidx         = nlist->shift;
 130     gid              = nlist->gid;
 131     shiftvec         = fr->shift_vec[0];
 132     fshift           = fr->fshift[0];
 133     facel            = _mm_set1_ps(fr->epsfac);
 134     charge           = mdatoms->chargeA;
 135     nvdwtype         = fr->ntype;
 136     vdwparam         = fr->nbfp;
 137     vdwtype          = mdatoms->typeA;
 138     vdwgridparam     = fr->ljpme_c6grid;
 139     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 140     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 141     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 142
 143     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 144     ewtab            = fr->ic->tabq_coul_FDV0;
 145     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 146     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 147
 148     /* Setup water-specific parameters */
 149     inr              = nlist->iinr[0];
 150     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 151     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 152     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 153     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 156     rcutoff_scalar   = fr->rcoulomb;
 157     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 158     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 159
 160     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 161     rvdw             = _mm_set1_ps(fr->rvdw);
 162
 163     /* Avoid stupid compiler warnings */
 164     jnrA = jnrB = jnrC = jnrD = 0;
 165     j_coord_offsetA = 0;
 166     j_coord_offsetB = 0;
 167     j_coord_offsetC = 0;
 168     j_coord_offsetD = 0;
 169
 170     outeriter        = 0;
 171     inneriter        = 0;
 172
 173     for(iidx=0;iidx<4*DIM;iidx++)
 174     {
 175         scratch[iidx] = 0.0;
 176     }
 177
 178     /* Start outer loop over neighborlists */
 179     for(iidx=0; iidx<nri; iidx++)
 180     {
 181         /* Load shift vector for this list */
 182         i_shift_offset   = DIM*shiftidx[iidx];
 183
 184         /* Load limits for loop over neighbors */
 185         j_index_start    = jindex[iidx];
 186         j_index_end      = jindex[iidx+1];
 187
 188         /* Get outer coordinate index */
 189         inr              = iinr[iidx];
 190         i_coord_offset   = DIM*inr;
 191
 192         /* Load i particle coords and add shift vector */
 193         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 194                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 195
 196         fix0             = _mm_setzero_ps();
 197         fiy0             = _mm_setzero_ps();
 198         fiz0             = _mm_setzero_ps();
 199         fix1             = _mm_setzero_ps();
 200         fiy1             = _mm_setzero_ps();
 201         fiz1             = _mm_setzero_ps();
 202         fix2             = _mm_setzero_ps();
 203         fiy2             = _mm_setzero_ps();
 204         fiz2             = _mm_setzero_ps();
 205         fix3             = _mm_setzero_ps();
 206         fiy3             = _mm_setzero_ps();
 207         fiz3             = _mm_setzero_ps();
 208
 209         /* Reset potential sums */
 210         velecsum         = _mm_setzero_ps();
 211         vvdwsum          = _mm_setzero_ps();
 212
 213         /* Start inner kernel loop */
 214         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 215         {
 216
 217             /* Get j neighbor index, and coordinate index */
 218             jnrA             = jjnr[jidx];
 219             jnrB             = jjnr[jidx+1];
 220             jnrC             = jjnr[jidx+2];
 221             jnrD             = jjnr[jidx+3];
 222             j_coord_offsetA  = DIM*jnrA;
 223             j_coord_offsetB  = DIM*jnrB;
 224             j_coord_offsetC  = DIM*jnrC;
 225             j_coord_offsetD  = DIM*jnrD;
 226
 227             /* load j atom coordinates */
 228             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 229                                               x+j_coord_offsetC,x+j_coord_offsetD,
 230                                               &jx0,&jy0,&jz0);
 231
 232             /* Calculate displacement vector */
 233             dx00             = _mm_sub_ps(ix0,jx0);
 234             dy00             = _mm_sub_ps(iy0,jy0);
 235             dz00             = _mm_sub_ps(iz0,jz0);
 236             dx10             = _mm_sub_ps(ix1,jx0);
 237             dy10             = _mm_sub_ps(iy1,jy0);
 238             dz10             = _mm_sub_ps(iz1,jz0);
 239             dx20             = _mm_sub_ps(ix2,jx0);
 240             dy20             = _mm_sub_ps(iy2,jy0);
 241             dz20             = _mm_sub_ps(iz2,jz0);
 242             dx30             = _mm_sub_ps(ix3,jx0);
 243             dy30             = _mm_sub_ps(iy3,jy0);
 244             dz30             = _mm_sub_ps(iz3,jz0);
 245
 246             /* Calculate squared distance and things based on it */
 247             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 248             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 249             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 250             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 251
 252             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 253             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 254             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 255             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 256
 257             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 258             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 259             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 260             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 261
 262             /* Load parameters for j particles */
 263             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 264                                                               charge+jnrC+0,charge+jnrD+0);
 265             vdwjidx0A        = 2*vdwtype[jnrA+0];
 266             vdwjidx0B        = 2*vdwtype[jnrB+0];
 267             vdwjidx0C        = 2*vdwtype[jnrC+0];
 268             vdwjidx0D        = 2*vdwtype[jnrD+0];
 269
 270             fjx0             = _mm_setzero_ps();
 271             fjy0             = _mm_setzero_ps();
 272             fjz0             = _mm_setzero_ps();
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             if (gmx_mm_any_lt(rsq00,rcutoff2))
 279             {
 280
 281             r00              = _mm_mul_ps(rsq00,rinv00);
 282
 283             /* Compute parameters for interactions between i and j atoms */
 284             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 285                                          vdwparam+vdwioffset0+vdwjidx0B,
 286                                          vdwparam+vdwioffset0+vdwjidx0C,
 287                                          vdwparam+vdwioffset0+vdwjidx0D,
 288                                          &c6_00,&c12_00);
 289             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 290                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 291                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 292                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 293
 294             /* Analytical LJ-PME */
 295             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 296             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 297             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 298             exponent         = gmx_simd_exp_r(ewcljrsq);
 299             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 300             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 301             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 302             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 303             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 304             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) ,
 305                                           _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));
 306             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 307             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);
 308
 309             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 313             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 314
 315             fscal            = fvdw;
 316
 317             fscal            = _mm_and_ps(fscal,cutoff_mask);
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm_mul_ps(fscal,dx00);
 321             ty               = _mm_mul_ps(fscal,dy00);
 322             tz               = _mm_mul_ps(fscal,dz00);
 323
 324             /* Update vectorial force */
 325             fix0             = _mm_add_ps(fix0,tx);
 326             fiy0             = _mm_add_ps(fiy0,ty);
 327             fiz0             = _mm_add_ps(fiz0,tz);
 328
 329             fjx0             = _mm_add_ps(fjx0,tx);
 330             fjy0             = _mm_add_ps(fjy0,ty);
 331             fjz0             = _mm_add_ps(fjz0,tz);
 332
 333             }
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             if (gmx_mm_any_lt(rsq10,rcutoff2))
 340             {
 341
 342             r10              = _mm_mul_ps(rsq10,rinv10);
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq10             = _mm_mul_ps(iq1,jq0);
 346
 347             /* EWALD ELECTROSTATICS */
 348
 349             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 350             ewrt             = _mm_mul_ps(r10,ewtabscale);
 351             ewitab           = _mm_cvttps_epi32(ewrt);
 352             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 353             ewitab           = _mm_slli_epi32(ewitab,2);
 354             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 355             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 356             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 357             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 358             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 359             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 360             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 361             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 362             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 363
 364             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 365
 366             /* Update potential sum for this i atom from the interaction with this j atom. */
 367             velec            = _mm_and_ps(velec,cutoff_mask);
 368             velecsum         = _mm_add_ps(velecsum,velec);
 369
 370             fscal            = felec;
 371
 372             fscal            = _mm_and_ps(fscal,cutoff_mask);
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm_mul_ps(fscal,dx10);
 376             ty               = _mm_mul_ps(fscal,dy10);
 377             tz               = _mm_mul_ps(fscal,dz10);
 378
 379             /* Update vectorial force */
 380             fix1             = _mm_add_ps(fix1,tx);
 381             fiy1             = _mm_add_ps(fiy1,ty);
 382             fiz1             = _mm_add_ps(fiz1,tz);
 383
 384             fjx0             = _mm_add_ps(fjx0,tx);
 385             fjy0             = _mm_add_ps(fjy0,ty);
 386             fjz0             = _mm_add_ps(fjz0,tz);
 387
 388             }
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             if (gmx_mm_any_lt(rsq20,rcutoff2))
 395             {
 396
 397             r20              = _mm_mul_ps(rsq20,rinv20);
 398
 399             /* Compute parameters for interactions between i and j atoms */
 400             qq20             = _mm_mul_ps(iq2,jq0);
 401
 402             /* EWALD ELECTROSTATICS */
 403
 404             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 405             ewrt             = _mm_mul_ps(r20,ewtabscale);
 406             ewitab           = _mm_cvttps_epi32(ewrt);
 407             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 408             ewitab           = _mm_slli_epi32(ewitab,2);
 409             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 410             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 411             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 412             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 413             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 414             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 415             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 416             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 417             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 418
 419             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 420
 421             /* Update potential sum for this i atom from the interaction with this j atom. */
 422             velec            = _mm_and_ps(velec,cutoff_mask);
 423             velecsum         = _mm_add_ps(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427             fscal            = _mm_and_ps(fscal,cutoff_mask);
 428
 429             /* Calculate temporary vectorial force */
 430             tx               = _mm_mul_ps(fscal,dx20);
 431             ty               = _mm_mul_ps(fscal,dy20);
 432             tz               = _mm_mul_ps(fscal,dz20);
 433
 434             /* Update vectorial force */
 435             fix2             = _mm_add_ps(fix2,tx);
 436             fiy2             = _mm_add_ps(fiy2,ty);
 437             fiz2             = _mm_add_ps(fiz2,tz);
 438
 439             fjx0             = _mm_add_ps(fjx0,tx);
 440             fjy0             = _mm_add_ps(fjy0,ty);
 441             fjz0             = _mm_add_ps(fjz0,tz);
 442
 443             }
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             if (gmx_mm_any_lt(rsq30,rcutoff2))
 450             {
 451
 452             r30              = _mm_mul_ps(rsq30,rinv30);
 453
 454             /* Compute parameters for interactions between i and j atoms */
 455             qq30             = _mm_mul_ps(iq3,jq0);
 456
 457             /* EWALD ELECTROSTATICS */
 458
 459             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 460             ewrt             = _mm_mul_ps(r30,ewtabscale);
 461             ewitab           = _mm_cvttps_epi32(ewrt);
 462             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 463             ewitab           = _mm_slli_epi32(ewitab,2);
 464             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 465             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 466             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 467             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 468             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 469             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 470             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 471             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 472             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 473
 474             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 475
 476             /* Update potential sum for this i atom from the interaction with this j atom. */
 477             velec            = _mm_and_ps(velec,cutoff_mask);
 478             velecsum         = _mm_add_ps(velecsum,velec);
 479
 480             fscal            = felec;
 481
 482             fscal            = _mm_and_ps(fscal,cutoff_mask);
 483
 484             /* Calculate temporary vectorial force */
 485             tx               = _mm_mul_ps(fscal,dx30);
 486             ty               = _mm_mul_ps(fscal,dy30);
 487             tz               = _mm_mul_ps(fscal,dz30);
 488
 489             /* Update vectorial force */
 490             fix3             = _mm_add_ps(fix3,tx);
 491             fiy3             = _mm_add_ps(fiy3,ty);
 492             fiz3             = _mm_add_ps(fiz3,tz);
 493
 494             fjx0             = _mm_add_ps(fjx0,tx);
 495             fjy0             = _mm_add_ps(fjy0,ty);
 496             fjz0             = _mm_add_ps(fjz0,tz);
 497
 498             }
 499
 500             fjptrA             = f+j_coord_offsetA;
 501             fjptrB             = f+j_coord_offsetB;
 502             fjptrC             = f+j_coord_offsetC;
 503             fjptrD             = f+j_coord_offsetD;
 504
 505             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 506
 507             /* Inner loop uses 200 flops */
 508         }
 509
 510         if(jidx<j_index_end)
 511         {
 512
 513             /* Get j neighbor index, and coordinate index */
 514             jnrlistA         = jjnr[jidx];
 515             jnrlistB         = jjnr[jidx+1];
 516             jnrlistC         = jjnr[jidx+2];
 517             jnrlistD         = jjnr[jidx+3];
 518             /* Sign of each element will be negative for non-real atoms.
 519              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 520              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 521              */
 522             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 523             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 524             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 525             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 526             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 527             j_coord_offsetA  = DIM*jnrA;
 528             j_coord_offsetB  = DIM*jnrB;
 529             j_coord_offsetC  = DIM*jnrC;
 530             j_coord_offsetD  = DIM*jnrD;
 531
 532             /* load j atom coordinates */
 533             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 534                                               x+j_coord_offsetC,x+j_coord_offsetD,
 535                                               &jx0,&jy0,&jz0);
 536
 537             /* Calculate displacement vector */
 538             dx00             = _mm_sub_ps(ix0,jx0);
 539             dy00             = _mm_sub_ps(iy0,jy0);
 540             dz00             = _mm_sub_ps(iz0,jz0);
 541             dx10             = _mm_sub_ps(ix1,jx0);
 542             dy10             = _mm_sub_ps(iy1,jy0);
 543             dz10             = _mm_sub_ps(iz1,jz0);
 544             dx20             = _mm_sub_ps(ix2,jx0);
 545             dy20             = _mm_sub_ps(iy2,jy0);
 546             dz20             = _mm_sub_ps(iz2,jz0);
 547             dx30             = _mm_sub_ps(ix3,jx0);
 548             dy30             = _mm_sub_ps(iy3,jy0);
 549             dz30             = _mm_sub_ps(iz3,jz0);
 550
 551             /* Calculate squared distance and things based on it */
 552             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 553             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 554             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 555             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 556
 557             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 558             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 559             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 560             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 561
 562             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 563             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 564             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 565             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 566
 567             /* Load parameters for j particles */
 568             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 569                                                               charge+jnrC+0,charge+jnrD+0);
 570             vdwjidx0A        = 2*vdwtype[jnrA+0];
 571             vdwjidx0B        = 2*vdwtype[jnrB+0];
 572             vdwjidx0C        = 2*vdwtype[jnrC+0];
 573             vdwjidx0D        = 2*vdwtype[jnrD+0];
 574
 575             fjx0             = _mm_setzero_ps();
 576             fjy0             = _mm_setzero_ps();
 577             fjz0             = _mm_setzero_ps();
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             if (gmx_mm_any_lt(rsq00,rcutoff2))
 584             {
 585
 586             r00              = _mm_mul_ps(rsq00,rinv00);
 587             r00              = _mm_andnot_ps(dummy_mask,r00);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 591                                          vdwparam+vdwioffset0+vdwjidx0B,
 592                                          vdwparam+vdwioffset0+vdwjidx0C,
 593                                          vdwparam+vdwioffset0+vdwjidx0D,
 594                                          &c6_00,&c12_00);
 595             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 596                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 597                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 598                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 599
 600             /* Analytical LJ-PME */
 601             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 602             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 603             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 604             exponent         = gmx_simd_exp_r(ewcljrsq);
 605             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 606             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 607             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 608             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 609             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 610             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) ,
 611                                           _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));
 612             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 613             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);
 614
 615             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 616
 617             /* Update potential sum for this i atom from the interaction with this j atom. */
 618             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 619             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 620             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 621
 622             fscal            = fvdw;
 623
 624             fscal            = _mm_and_ps(fscal,cutoff_mask);
 625
 626             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 627
 628             /* Calculate temporary vectorial force */
 629             tx               = _mm_mul_ps(fscal,dx00);
 630             ty               = _mm_mul_ps(fscal,dy00);
 631             tz               = _mm_mul_ps(fscal,dz00);
 632
 633             /* Update vectorial force */
 634             fix0             = _mm_add_ps(fix0,tx);
 635             fiy0             = _mm_add_ps(fiy0,ty);
 636             fiz0             = _mm_add_ps(fiz0,tz);
 637
 638             fjx0             = _mm_add_ps(fjx0,tx);
 639             fjy0             = _mm_add_ps(fjy0,ty);
 640             fjz0             = _mm_add_ps(fjz0,tz);
 641
 642             }
 643
 644             /**************************
 645              * CALCULATE INTERACTIONS *
 646              **************************/
 647
 648             if (gmx_mm_any_lt(rsq10,rcutoff2))
 649             {
 650
 651             r10              = _mm_mul_ps(rsq10,rinv10);
 652             r10              = _mm_andnot_ps(dummy_mask,r10);
 653
 654             /* Compute parameters for interactions between i and j atoms */
 655             qq10             = _mm_mul_ps(iq1,jq0);
 656
 657             /* EWALD ELECTROSTATICS */
 658
 659             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 660             ewrt             = _mm_mul_ps(r10,ewtabscale);
 661             ewitab           = _mm_cvttps_epi32(ewrt);
 662             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 663             ewitab           = _mm_slli_epi32(ewitab,2);
 664             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 665             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 666             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 667             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 668             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 669             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 670             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 671             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 672             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 673
 674             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 675
 676             /* Update potential sum for this i atom from the interaction with this j atom. */
 677             velec            = _mm_and_ps(velec,cutoff_mask);
 678             velec            = _mm_andnot_ps(dummy_mask,velec);
 679             velecsum         = _mm_add_ps(velecsum,velec);
 680
 681             fscal            = felec;
 682
 683             fscal            = _mm_and_ps(fscal,cutoff_mask);
 684
 685             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 686
 687             /* Calculate temporary vectorial force */
 688             tx               = _mm_mul_ps(fscal,dx10);
 689             ty               = _mm_mul_ps(fscal,dy10);
 690             tz               = _mm_mul_ps(fscal,dz10);
 691
 692             /* Update vectorial force */
 693             fix1             = _mm_add_ps(fix1,tx);
 694             fiy1             = _mm_add_ps(fiy1,ty);
 695             fiz1             = _mm_add_ps(fiz1,tz);
 696
 697             fjx0             = _mm_add_ps(fjx0,tx);
 698             fjy0             = _mm_add_ps(fjy0,ty);
 699             fjz0             = _mm_add_ps(fjz0,tz);
 700
 701             }
 702
 703             /**************************
 704              * CALCULATE INTERACTIONS *
 705              **************************/
 706
 707             if (gmx_mm_any_lt(rsq20,rcutoff2))
 708             {
 709
 710             r20              = _mm_mul_ps(rsq20,rinv20);
 711             r20              = _mm_andnot_ps(dummy_mask,r20);
 712
 713             /* Compute parameters for interactions between i and j atoms */
 714             qq20             = _mm_mul_ps(iq2,jq0);
 715
 716             /* EWALD ELECTROSTATICS */
 717
 718             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 719             ewrt             = _mm_mul_ps(r20,ewtabscale);
 720             ewitab           = _mm_cvttps_epi32(ewrt);
 721             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 722             ewitab           = _mm_slli_epi32(ewitab,2);
 723             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 724             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 725             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 726             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 727             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 728             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 729             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 730             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 731             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 732
 733             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 734
 735             /* Update potential sum for this i atom from the interaction with this j atom. */
 736             velec            = _mm_and_ps(velec,cutoff_mask);
 737             velec            = _mm_andnot_ps(dummy_mask,velec);
 738             velecsum         = _mm_add_ps(velecsum,velec);
 739
 740             fscal            = felec;
 741
 742             fscal            = _mm_and_ps(fscal,cutoff_mask);
 743
 744             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 745
 746             /* Calculate temporary vectorial force */
 747             tx               = _mm_mul_ps(fscal,dx20);
 748             ty               = _mm_mul_ps(fscal,dy20);
 749             tz               = _mm_mul_ps(fscal,dz20);
 750
 751             /* Update vectorial force */
 752             fix2             = _mm_add_ps(fix2,tx);
 753             fiy2             = _mm_add_ps(fiy2,ty);
 754             fiz2             = _mm_add_ps(fiz2,tz);
 755
 756             fjx0             = _mm_add_ps(fjx0,tx);
 757             fjy0             = _mm_add_ps(fjy0,ty);
 758             fjz0             = _mm_add_ps(fjz0,tz);
 759
 760             }
 761
 762             /**************************
 763              * CALCULATE INTERACTIONS *
 764              **************************/
 765
 766             if (gmx_mm_any_lt(rsq30,rcutoff2))
 767             {
 768
 769             r30              = _mm_mul_ps(rsq30,rinv30);
 770             r30              = _mm_andnot_ps(dummy_mask,r30);
 771
 772             /* Compute parameters for interactions between i and j atoms */
 773             qq30             = _mm_mul_ps(iq3,jq0);
 774
 775             /* EWALD ELECTROSTATICS */
 776
 777             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 778             ewrt             = _mm_mul_ps(r30,ewtabscale);
 779             ewitab           = _mm_cvttps_epi32(ewrt);
 780             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 781             ewitab           = _mm_slli_epi32(ewitab,2);
 782             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 783             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 784             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 785             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 786             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 787             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 788             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 789             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 790             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 791
 792             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 793
 794             /* Update potential sum for this i atom from the interaction with this j atom. */
 795             velec            = _mm_and_ps(velec,cutoff_mask);
 796             velec            = _mm_andnot_ps(dummy_mask,velec);
 797             velecsum         = _mm_add_ps(velecsum,velec);
 798
 799             fscal            = felec;
 800
 801             fscal            = _mm_and_ps(fscal,cutoff_mask);
 802
 803             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 804
 805             /* Calculate temporary vectorial force */
 806             tx               = _mm_mul_ps(fscal,dx30);
 807             ty               = _mm_mul_ps(fscal,dy30);
 808             tz               = _mm_mul_ps(fscal,dz30);
 809
 810             /* Update vectorial force */
 811             fix3             = _mm_add_ps(fix3,tx);
 812             fiy3             = _mm_add_ps(fiy3,ty);
 813             fiz3             = _mm_add_ps(fiz3,tz);
 814
 815             fjx0             = _mm_add_ps(fjx0,tx);
 816             fjy0             = _mm_add_ps(fjy0,ty);
 817             fjz0             = _mm_add_ps(fjz0,tz);
 818
 819             }
 820
 821             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 822             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 823             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 824             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 825
 826             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 827
 828             /* Inner loop uses 204 flops */
 829         }
 830
 831         /* End of innermost loop */
 832
 833         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 834                                               f+i_coord_offset,fshift+i_shift_offset);
 835
 836         ggid                        = gid[iidx];
 837         /* Update potential energies */
 838         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 839         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 840
 841         /* Increment number of inner iterations */
 842         inneriter                  += j_index_end - j_index_start;
 843
 844         /* Outer loop uses 26 flops */
 845     }
 846
 847     /* Increment number of outer iterations */
 848     outeriter        += nri;
 849
 850     /* Update outer/inner flops */
 851
 852     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*204);
 853 }
 854 /*
 855  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_single
 856  * Electrostatics interaction: Ewald
 857  * VdW interaction:            LJEwald
 858  * Geometry:                   Water4-Particle
 859  * Calculate force/pot:        Force
 860  */
 861 void
 862 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_single
 863                     (t_nblist                    * gmx_restrict       nlist,
 864                      rvec                        * gmx_restrict          xx,
 865                      rvec                        * gmx_restrict          ff,
 866                      t_forcerec                  * gmx_restrict          fr,
 867                      t_mdatoms                   * gmx_restrict     mdatoms,
 868                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 869                      t_nrnb                      * gmx_restrict        nrnb)
 870 {
 871     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 872      * just 0 for non-waters.
 873      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 874      * jnr indices corresponding to data put in the four positions in the SIMD register.
 875      */
 876     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 877     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 878     int              jnrA,jnrB,jnrC,jnrD;
 879     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 880     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 881     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 882     real             rcutoff_scalar;
 883     real             *shiftvec,*fshift,*x,*f;
 884     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 885     real             scratch[4*DIM];
 886     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 887     int              vdwioffset0;
 888     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 889     int              vdwioffset1;
 890     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 891     int              vdwioffset2;
 892     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 893     int              vdwioffset3;
 894     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 895     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 896     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 897     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 898     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 899     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 900     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 901     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 902     real             *charge;
 903     int              nvdwtype;
 904     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 905     int              *vdwtype;
 906     real             *vdwparam;
 907     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 908     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 909     __m128           c6grid_00;
 910     __m128           c6grid_10;
 911     __m128           c6grid_20;
 912     __m128           c6grid_30;
 913     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 914     real             *vdwgridparam;
 915     __m128           one_half = _mm_set1_ps(0.5);
 916     __m128           minus_one = _mm_set1_ps(-1.0);
 917     __m128i          ewitab;
 918     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 919     real             *ewtab;
 920     __m128           dummy_mask,cutoff_mask;
 921     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 922     __m128           one     = _mm_set1_ps(1.0);
 923     __m128           two     = _mm_set1_ps(2.0);
 924     x                = xx[0];
 925     f                = ff[0];
 926
 927     nri              = nlist->nri;
 928     iinr             = nlist->iinr;
 929     jindex           = nlist->jindex;
 930     jjnr             = nlist->jjnr;
 931     shiftidx         = nlist->shift;
 932     gid              = nlist->gid;
 933     shiftvec         = fr->shift_vec[0];
 934     fshift           = fr->fshift[0];
 935     facel            = _mm_set1_ps(fr->epsfac);
 936     charge           = mdatoms->chargeA;
 937     nvdwtype         = fr->ntype;
 938     vdwparam         = fr->nbfp;
 939     vdwtype          = mdatoms->typeA;
 940     vdwgridparam     = fr->ljpme_c6grid;
 941     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 942     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 943     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 944
 945     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 946     ewtab            = fr->ic->tabq_coul_F;
 947     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 948     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 949
 950     /* Setup water-specific parameters */
 951     inr              = nlist->iinr[0];
 952     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 953     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 954     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 955     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 956
 957     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 958     rcutoff_scalar   = fr->rcoulomb;
 959     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 960     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 961
 962     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 963     rvdw             = _mm_set1_ps(fr->rvdw);
 964
 965     /* Avoid stupid compiler warnings */
 966     jnrA = jnrB = jnrC = jnrD = 0;
 967     j_coord_offsetA = 0;
 968     j_coord_offsetB = 0;
 969     j_coord_offsetC = 0;
 970     j_coord_offsetD = 0;
 971
 972     outeriter        = 0;
 973     inneriter        = 0;
 974
 975     for(iidx=0;iidx<4*DIM;iidx++)
 976     {
 977         scratch[iidx] = 0.0;
 978     }
 979
 980     /* Start outer loop over neighborlists */
 981     for(iidx=0; iidx<nri; iidx++)
 982     {
 983         /* Load shift vector for this list */
 984         i_shift_offset   = DIM*shiftidx[iidx];
 985
 986         /* Load limits for loop over neighbors */
 987         j_index_start    = jindex[iidx];
 988         j_index_end      = jindex[iidx+1];
 989
 990         /* Get outer coordinate index */
 991         inr              = iinr[iidx];
 992         i_coord_offset   = DIM*inr;
 993
 994         /* Load i particle coords and add shift vector */
 995         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 996                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 997
 998         fix0             = _mm_setzero_ps();
 999         fiy0             = _mm_setzero_ps();
1000         fiz0             = _mm_setzero_ps();
1001         fix1             = _mm_setzero_ps();
1002         fiy1             = _mm_setzero_ps();
1003         fiz1             = _mm_setzero_ps();
1004         fix2             = _mm_setzero_ps();
1005         fiy2             = _mm_setzero_ps();
1006         fiz2             = _mm_setzero_ps();
1007         fix3             = _mm_setzero_ps();
1008         fiy3             = _mm_setzero_ps();
1009         fiz3             = _mm_setzero_ps();
1010
1011         /* Start inner kernel loop */
1012         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1013         {
1014
1015             /* Get j neighbor index, and coordinate index */
1016             jnrA             = jjnr[jidx];
1017             jnrB             = jjnr[jidx+1];
1018             jnrC             = jjnr[jidx+2];
1019             jnrD             = jjnr[jidx+3];
1020             j_coord_offsetA  = DIM*jnrA;
1021             j_coord_offsetB  = DIM*jnrB;
1022             j_coord_offsetC  = DIM*jnrC;
1023             j_coord_offsetD  = DIM*jnrD;
1024
1025             /* load j atom coordinates */
1026             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1027                                               x+j_coord_offsetC,x+j_coord_offsetD,
1028                                               &jx0,&jy0,&jz0);
1029
1030             /* Calculate displacement vector */
1031             dx00             = _mm_sub_ps(ix0,jx0);
1032             dy00             = _mm_sub_ps(iy0,jy0);
1033             dz00             = _mm_sub_ps(iz0,jz0);
1034             dx10             = _mm_sub_ps(ix1,jx0);
1035             dy10             = _mm_sub_ps(iy1,jy0);
1036             dz10             = _mm_sub_ps(iz1,jz0);
1037             dx20             = _mm_sub_ps(ix2,jx0);
1038             dy20             = _mm_sub_ps(iy2,jy0);
1039             dz20             = _mm_sub_ps(iz2,jz0);
1040             dx30             = _mm_sub_ps(ix3,jx0);
1041             dy30             = _mm_sub_ps(iy3,jy0);
1042             dz30             = _mm_sub_ps(iz3,jz0);
1043
1044             /* Calculate squared distance and things based on it */
1045             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1046             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1047             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1048             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1049
1050             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1051             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1052             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1053             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1054
1055             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1056             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1057             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1058             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1059
1060             /* Load parameters for j particles */
1061             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1062                                                               charge+jnrC+0,charge+jnrD+0);
1063             vdwjidx0A        = 2*vdwtype[jnrA+0];
1064             vdwjidx0B        = 2*vdwtype[jnrB+0];
1065             vdwjidx0C        = 2*vdwtype[jnrC+0];
1066             vdwjidx0D        = 2*vdwtype[jnrD+0];
1067
1068             fjx0             = _mm_setzero_ps();
1069             fjy0             = _mm_setzero_ps();
1070             fjz0             = _mm_setzero_ps();
1071
1072             /**************************
1073              * CALCULATE INTERACTIONS *
1074              **************************/
1075
1076             if (gmx_mm_any_lt(rsq00,rcutoff2))
1077             {
1078
1079             r00              = _mm_mul_ps(rsq00,rinv00);
1080
1081             /* Compute parameters for interactions between i and j atoms */
1082             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1083                                          vdwparam+vdwioffset0+vdwjidx0B,
1084                                          vdwparam+vdwioffset0+vdwjidx0C,
1085                                          vdwparam+vdwioffset0+vdwjidx0D,
1086                                          &c6_00,&c12_00);
1087             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1088                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1089                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1090                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1091
1092             /* Analytical LJ-PME */
1093             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1094             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1095             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1096             exponent         = gmx_simd_exp_r(ewcljrsq);
1097             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1098             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1099             /* f6A = 6 * C6grid * (1 - poly) */
1100             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1101             /* f6B = C6grid * exponent * beta^6 */
1102             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1103             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1104             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);
1105
1106             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1107
1108             fscal            = fvdw;
1109
1110             fscal            = _mm_and_ps(fscal,cutoff_mask);
1111
1112             /* Calculate temporary vectorial force */
1113             tx               = _mm_mul_ps(fscal,dx00);
1114             ty               = _mm_mul_ps(fscal,dy00);
1115             tz               = _mm_mul_ps(fscal,dz00);
1116
1117             /* Update vectorial force */
1118             fix0             = _mm_add_ps(fix0,tx);
1119             fiy0             = _mm_add_ps(fiy0,ty);
1120             fiz0             = _mm_add_ps(fiz0,tz);
1121
1122             fjx0             = _mm_add_ps(fjx0,tx);
1123             fjy0             = _mm_add_ps(fjy0,ty);
1124             fjz0             = _mm_add_ps(fjz0,tz);
1125
1126             }
1127
1128             /**************************
1129              * CALCULATE INTERACTIONS *
1130              **************************/
1131
1132             if (gmx_mm_any_lt(rsq10,rcutoff2))
1133             {
1134
1135             r10              = _mm_mul_ps(rsq10,rinv10);
1136
1137             /* Compute parameters for interactions between i and j atoms */
1138             qq10             = _mm_mul_ps(iq1,jq0);
1139
1140             /* EWALD ELECTROSTATICS */
1141
1142             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1143             ewrt             = _mm_mul_ps(r10,ewtabscale);
1144             ewitab           = _mm_cvttps_epi32(ewrt);
1145             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1146             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1147                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1148                                          &ewtabF,&ewtabFn);
1149             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1150             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1151
1152             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1153
1154             fscal            = felec;
1155
1156             fscal            = _mm_and_ps(fscal,cutoff_mask);
1157
1158             /* Calculate temporary vectorial force */
1159             tx               = _mm_mul_ps(fscal,dx10);
1160             ty               = _mm_mul_ps(fscal,dy10);
1161             tz               = _mm_mul_ps(fscal,dz10);
1162
1163             /* Update vectorial force */
1164             fix1             = _mm_add_ps(fix1,tx);
1165             fiy1             = _mm_add_ps(fiy1,ty);
1166             fiz1             = _mm_add_ps(fiz1,tz);
1167
1168             fjx0             = _mm_add_ps(fjx0,tx);
1169             fjy0             = _mm_add_ps(fjy0,ty);
1170             fjz0             = _mm_add_ps(fjz0,tz);
1171
1172             }
1173
1174             /**************************
1175              * CALCULATE INTERACTIONS *
1176              **************************/
1177
1178             if (gmx_mm_any_lt(rsq20,rcutoff2))
1179             {
1180
1181             r20              = _mm_mul_ps(rsq20,rinv20);
1182
1183             /* Compute parameters for interactions between i and j atoms */
1184             qq20             = _mm_mul_ps(iq2,jq0);
1185
1186             /* EWALD ELECTROSTATICS */
1187
1188             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1189             ewrt             = _mm_mul_ps(r20,ewtabscale);
1190             ewitab           = _mm_cvttps_epi32(ewrt);
1191             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1192             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1193                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1194                                          &ewtabF,&ewtabFn);
1195             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1196             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1197
1198             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1199
1200             fscal            = felec;
1201
1202             fscal            = _mm_and_ps(fscal,cutoff_mask);
1203
1204             /* Calculate temporary vectorial force */
1205             tx               = _mm_mul_ps(fscal,dx20);
1206             ty               = _mm_mul_ps(fscal,dy20);
1207             tz               = _mm_mul_ps(fscal,dz20);
1208
1209             /* Update vectorial force */
1210             fix2             = _mm_add_ps(fix2,tx);
1211             fiy2             = _mm_add_ps(fiy2,ty);
1212             fiz2             = _mm_add_ps(fiz2,tz);
1213
1214             fjx0             = _mm_add_ps(fjx0,tx);
1215             fjy0             = _mm_add_ps(fjy0,ty);
1216             fjz0             = _mm_add_ps(fjz0,tz);
1217
1218             }
1219
1220             /**************************
1221              * CALCULATE INTERACTIONS *
1222              **************************/
1223
1224             if (gmx_mm_any_lt(rsq30,rcutoff2))
1225             {
1226
1227             r30              = _mm_mul_ps(rsq30,rinv30);
1228
1229             /* Compute parameters for interactions between i and j atoms */
1230             qq30             = _mm_mul_ps(iq3,jq0);
1231
1232             /* EWALD ELECTROSTATICS */
1233
1234             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1235             ewrt             = _mm_mul_ps(r30,ewtabscale);
1236             ewitab           = _mm_cvttps_epi32(ewrt);
1237             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1238             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1239                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1240                                          &ewtabF,&ewtabFn);
1241             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1242             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1243
1244             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1245
1246             fscal            = felec;
1247
1248             fscal            = _mm_and_ps(fscal,cutoff_mask);
1249
1250             /* Calculate temporary vectorial force */
1251             tx               = _mm_mul_ps(fscal,dx30);
1252             ty               = _mm_mul_ps(fscal,dy30);
1253             tz               = _mm_mul_ps(fscal,dz30);
1254
1255             /* Update vectorial force */
1256             fix3             = _mm_add_ps(fix3,tx);
1257             fiy3             = _mm_add_ps(fiy3,ty);
1258             fiz3             = _mm_add_ps(fiz3,tz);
1259
1260             fjx0             = _mm_add_ps(fjx0,tx);
1261             fjy0             = _mm_add_ps(fjy0,ty);
1262             fjz0             = _mm_add_ps(fjz0,tz);
1263
1264             }
1265
1266             fjptrA             = f+j_coord_offsetA;
1267             fjptrB             = f+j_coord_offsetB;
1268             fjptrC             = f+j_coord_offsetC;
1269             fjptrD             = f+j_coord_offsetD;
1270
1271             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1272
1273             /* Inner loop uses 166 flops */
1274         }
1275
1276         if(jidx<j_index_end)
1277         {
1278
1279             /* Get j neighbor index, and coordinate index */
1280             jnrlistA         = jjnr[jidx];
1281             jnrlistB         = jjnr[jidx+1];
1282             jnrlistC         = jjnr[jidx+2];
1283             jnrlistD         = jjnr[jidx+3];
1284             /* Sign of each element will be negative for non-real atoms.
1285              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1286              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1287              */
1288             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1289             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1290             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1291             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1292             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1293             j_coord_offsetA  = DIM*jnrA;
1294             j_coord_offsetB  = DIM*jnrB;
1295             j_coord_offsetC  = DIM*jnrC;
1296             j_coord_offsetD  = DIM*jnrD;
1297
1298             /* load j atom coordinates */
1299             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1300                                               x+j_coord_offsetC,x+j_coord_offsetD,
1301                                               &jx0,&jy0,&jz0);
1302
1303             /* Calculate displacement vector */
1304             dx00             = _mm_sub_ps(ix0,jx0);
1305             dy00             = _mm_sub_ps(iy0,jy0);
1306             dz00             = _mm_sub_ps(iz0,jz0);
1307             dx10             = _mm_sub_ps(ix1,jx0);
1308             dy10             = _mm_sub_ps(iy1,jy0);
1309             dz10             = _mm_sub_ps(iz1,jz0);
1310             dx20             = _mm_sub_ps(ix2,jx0);
1311             dy20             = _mm_sub_ps(iy2,jy0);
1312             dz20             = _mm_sub_ps(iz2,jz0);
1313             dx30             = _mm_sub_ps(ix3,jx0);
1314             dy30             = _mm_sub_ps(iy3,jy0);
1315             dz30             = _mm_sub_ps(iz3,jz0);
1316
1317             /* Calculate squared distance and things based on it */
1318             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1319             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1320             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1321             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1322
1323             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1324             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1325             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1326             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1327
1328             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1329             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1330             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1331             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1332
1333             /* Load parameters for j particles */
1334             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1335                                                               charge+jnrC+0,charge+jnrD+0);
1336             vdwjidx0A        = 2*vdwtype[jnrA+0];
1337             vdwjidx0B        = 2*vdwtype[jnrB+0];
1338             vdwjidx0C        = 2*vdwtype[jnrC+0];
1339             vdwjidx0D        = 2*vdwtype[jnrD+0];
1340
1341             fjx0             = _mm_setzero_ps();
1342             fjy0             = _mm_setzero_ps();
1343             fjz0             = _mm_setzero_ps();
1344
1345             /**************************
1346              * CALCULATE INTERACTIONS *
1347              **************************/
1348
1349             if (gmx_mm_any_lt(rsq00,rcutoff2))
1350             {
1351
1352             r00              = _mm_mul_ps(rsq00,rinv00);
1353             r00              = _mm_andnot_ps(dummy_mask,r00);
1354
1355             /* Compute parameters for interactions between i and j atoms */
1356             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1357                                          vdwparam+vdwioffset0+vdwjidx0B,
1358                                          vdwparam+vdwioffset0+vdwjidx0C,
1359                                          vdwparam+vdwioffset0+vdwjidx0D,
1360                                          &c6_00,&c12_00);
1361             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1362                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1363                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1364                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1365
1366             /* Analytical LJ-PME */
1367             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1368             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1369             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1370             exponent         = gmx_simd_exp_r(ewcljrsq);
1371             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1372             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1373             /* f6A = 6 * C6grid * (1 - poly) */
1374             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1375             /* f6B = C6grid * exponent * beta^6 */
1376             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1377             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1378             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);
1379
1380             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1381
1382             fscal            = fvdw;
1383
1384             fscal            = _mm_and_ps(fscal,cutoff_mask);
1385
1386             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1387
1388             /* Calculate temporary vectorial force */
1389             tx               = _mm_mul_ps(fscal,dx00);
1390             ty               = _mm_mul_ps(fscal,dy00);
1391             tz               = _mm_mul_ps(fscal,dz00);
1392
1393             /* Update vectorial force */
1394             fix0             = _mm_add_ps(fix0,tx);
1395             fiy0             = _mm_add_ps(fiy0,ty);
1396             fiz0             = _mm_add_ps(fiz0,tz);
1397
1398             fjx0             = _mm_add_ps(fjx0,tx);
1399             fjy0             = _mm_add_ps(fjy0,ty);
1400             fjz0             = _mm_add_ps(fjz0,tz);
1401
1402             }
1403
1404             /**************************
1405              * CALCULATE INTERACTIONS *
1406              **************************/
1407
1408             if (gmx_mm_any_lt(rsq10,rcutoff2))
1409             {
1410
1411             r10              = _mm_mul_ps(rsq10,rinv10);
1412             r10              = _mm_andnot_ps(dummy_mask,r10);
1413
1414             /* Compute parameters for interactions between i and j atoms */
1415             qq10             = _mm_mul_ps(iq1,jq0);
1416
1417             /* EWALD ELECTROSTATICS */
1418
1419             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1420             ewrt             = _mm_mul_ps(r10,ewtabscale);
1421             ewitab           = _mm_cvttps_epi32(ewrt);
1422             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1423             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1424                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1425                                          &ewtabF,&ewtabFn);
1426             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1427             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1428
1429             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1430
1431             fscal            = felec;
1432
1433             fscal            = _mm_and_ps(fscal,cutoff_mask);
1434
1435             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1436
1437             /* Calculate temporary vectorial force */
1438             tx               = _mm_mul_ps(fscal,dx10);
1439             ty               = _mm_mul_ps(fscal,dy10);
1440             tz               = _mm_mul_ps(fscal,dz10);
1441
1442             /* Update vectorial force */
1443             fix1             = _mm_add_ps(fix1,tx);
1444             fiy1             = _mm_add_ps(fiy1,ty);
1445             fiz1             = _mm_add_ps(fiz1,tz);
1446
1447             fjx0             = _mm_add_ps(fjx0,tx);
1448             fjy0             = _mm_add_ps(fjy0,ty);
1449             fjz0             = _mm_add_ps(fjz0,tz);
1450
1451             }
1452
1453             /**************************
1454              * CALCULATE INTERACTIONS *
1455              **************************/
1456
1457             if (gmx_mm_any_lt(rsq20,rcutoff2))
1458             {
1459
1460             r20              = _mm_mul_ps(rsq20,rinv20);
1461             r20              = _mm_andnot_ps(dummy_mask,r20);
1462
1463             /* Compute parameters for interactions between i and j atoms */
1464             qq20             = _mm_mul_ps(iq2,jq0);
1465
1466             /* EWALD ELECTROSTATICS */
1467
1468             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1469             ewrt             = _mm_mul_ps(r20,ewtabscale);
1470             ewitab           = _mm_cvttps_epi32(ewrt);
1471             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1472             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1473                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1474                                          &ewtabF,&ewtabFn);
1475             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1476             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1477
1478             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1479
1480             fscal            = felec;
1481
1482             fscal            = _mm_and_ps(fscal,cutoff_mask);
1483
1484             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1485
1486             /* Calculate temporary vectorial force */
1487             tx               = _mm_mul_ps(fscal,dx20);
1488             ty               = _mm_mul_ps(fscal,dy20);
1489             tz               = _mm_mul_ps(fscal,dz20);
1490
1491             /* Update vectorial force */
1492             fix2             = _mm_add_ps(fix2,tx);
1493             fiy2             = _mm_add_ps(fiy2,ty);
1494             fiz2             = _mm_add_ps(fiz2,tz);
1495
1496             fjx0             = _mm_add_ps(fjx0,tx);
1497             fjy0             = _mm_add_ps(fjy0,ty);
1498             fjz0             = _mm_add_ps(fjz0,tz);
1499
1500             }
1501
1502             /**************************
1503              * CALCULATE INTERACTIONS *
1504              **************************/
1505
1506             if (gmx_mm_any_lt(rsq30,rcutoff2))
1507             {
1508
1509             r30              = _mm_mul_ps(rsq30,rinv30);
1510             r30              = _mm_andnot_ps(dummy_mask,r30);
1511
1512             /* Compute parameters for interactions between i and j atoms */
1513             qq30             = _mm_mul_ps(iq3,jq0);
1514
1515             /* EWALD ELECTROSTATICS */
1516
1517             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1518             ewrt             = _mm_mul_ps(r30,ewtabscale);
1519             ewitab           = _mm_cvttps_epi32(ewrt);
1520             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1521             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1522                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1523                                          &ewtabF,&ewtabFn);
1524             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1525             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1526
1527             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1528
1529             fscal            = felec;
1530
1531             fscal            = _mm_and_ps(fscal,cutoff_mask);
1532
1533             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1534
1535             /* Calculate temporary vectorial force */
1536             tx               = _mm_mul_ps(fscal,dx30);
1537             ty               = _mm_mul_ps(fscal,dy30);
1538             tz               = _mm_mul_ps(fscal,dz30);
1539
1540             /* Update vectorial force */
1541             fix3             = _mm_add_ps(fix3,tx);
1542             fiy3             = _mm_add_ps(fiy3,ty);
1543             fiz3             = _mm_add_ps(fiz3,tz);
1544
1545             fjx0             = _mm_add_ps(fjx0,tx);
1546             fjy0             = _mm_add_ps(fjy0,ty);
1547             fjz0             = _mm_add_ps(fjz0,tz);
1548
1549             }
1550
1551             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1552             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1553             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1554             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1555
1556             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1557
1558             /* Inner loop uses 170 flops */
1559         }
1560
1561         /* End of innermost loop */
1562
1563         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1564                                               f+i_coord_offset,fshift+i_shift_offset);
1565
1566         /* Increment number of inner iterations */
1567         inneriter                  += j_index_end - j_index_start;
1568
1569         /* Outer loop uses 24 flops */
1570     }
1571
1572     /* Increment number of outer iterations */
1573     outeriter        += nri;
1574
1575     /* Update outer/inner flops */
1576
1577     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*170);
1578 }