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) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse2_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128           c6grid_00;
 105     __m128           c6grid_10;
 106     __m128           c6grid_20;
 107     __m128           c6grid_30;
 108     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 109     real             *vdwgridparam;
 110     __m128           one_half = _mm_set1_ps(0.5);
 111     __m128           minus_one = _mm_set1_ps(-1.0);
 112     __m128i          ewitab;
 113     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     real             *ewtab;
 115     __m128           dummy_mask,cutoff_mask;
 116     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 117     __m128           one     = _mm_set1_ps(1.0);
 118     __m128           two     = _mm_set1_ps(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_ps(fr->ic->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135     vdwgridparam     = fr->ljpme_c6grid;
 136     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 137     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 138     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 139
 140     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 141     ewtab            = fr->ic->tabq_coul_FDV0;
 142     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 143     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 148     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 149     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 150     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 151
 152     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 153     rcutoff_scalar   = fr->ic->rcoulomb;
 154     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 155     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 156
 157     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 158     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 159
 160     /* Avoid stupid compiler warnings */
 161     jnrA = jnrB = jnrC = jnrD = 0;
 162     j_coord_offsetA = 0;
 163     j_coord_offsetB = 0;
 164     j_coord_offsetC = 0;
 165     j_coord_offsetD = 0;
 166
 167     outeriter        = 0;
 168     inneriter        = 0;
 169
 170     for(iidx=0;iidx<4*DIM;iidx++)
 171     {
 172         scratch[iidx] = 0.0;
 173     }
 174
 175     /* Start outer loop over neighborlists */
 176     for(iidx=0; iidx<nri; iidx++)
 177     {
 178         /* Load shift vector for this list */
 179         i_shift_offset   = DIM*shiftidx[iidx];
 180
 181         /* Load limits for loop over neighbors */
 182         j_index_start    = jindex[iidx];
 183         j_index_end      = jindex[iidx+1];
 184
 185         /* Get outer coordinate index */
 186         inr              = iinr[iidx];
 187         i_coord_offset   = DIM*inr;
 188
 189         /* Load i particle coords and add shift vector */
 190         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 191                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 192
 193         fix0             = _mm_setzero_ps();
 194         fiy0             = _mm_setzero_ps();
 195         fiz0             = _mm_setzero_ps();
 196         fix1             = _mm_setzero_ps();
 197         fiy1             = _mm_setzero_ps();
 198         fiz1             = _mm_setzero_ps();
 199         fix2             = _mm_setzero_ps();
 200         fiy2             = _mm_setzero_ps();
 201         fiz2             = _mm_setzero_ps();
 202         fix3             = _mm_setzero_ps();
 203         fiy3             = _mm_setzero_ps();
 204         fiz3             = _mm_setzero_ps();
 205
 206         /* Reset potential sums */
 207         velecsum         = _mm_setzero_ps();
 208         vvdwsum          = _mm_setzero_ps();
 209
 210         /* Start inner kernel loop */
 211         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 212         {
 213
 214             /* Get j neighbor index, and coordinate index */
 215             jnrA             = jjnr[jidx];
 216             jnrB             = jjnr[jidx+1];
 217             jnrC             = jjnr[jidx+2];
 218             jnrD             = jjnr[jidx+3];
 219             j_coord_offsetA  = DIM*jnrA;
 220             j_coord_offsetB  = DIM*jnrB;
 221             j_coord_offsetC  = DIM*jnrC;
 222             j_coord_offsetD  = DIM*jnrD;
 223
 224             /* load j atom coordinates */
 225             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 226                                               x+j_coord_offsetC,x+j_coord_offsetD,
 227                                               &jx0,&jy0,&jz0);
 228
 229             /* Calculate displacement vector */
 230             dx00             = _mm_sub_ps(ix0,jx0);
 231             dy00             = _mm_sub_ps(iy0,jy0);
 232             dz00             = _mm_sub_ps(iz0,jz0);
 233             dx10             = _mm_sub_ps(ix1,jx0);
 234             dy10             = _mm_sub_ps(iy1,jy0);
 235             dz10             = _mm_sub_ps(iz1,jz0);
 236             dx20             = _mm_sub_ps(ix2,jx0);
 237             dy20             = _mm_sub_ps(iy2,jy0);
 238             dz20             = _mm_sub_ps(iz2,jz0);
 239             dx30             = _mm_sub_ps(ix3,jx0);
 240             dy30             = _mm_sub_ps(iy3,jy0);
 241             dz30             = _mm_sub_ps(iz3,jz0);
 242
 243             /* Calculate squared distance and things based on it */
 244             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 245             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 246             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 247             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 248
 249             rinv00           = sse2_invsqrt_f(rsq00);
 250             rinv10           = sse2_invsqrt_f(rsq10);
 251             rinv20           = sse2_invsqrt_f(rsq20);
 252             rinv30           = sse2_invsqrt_f(rsq30);
 253
 254             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 255             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 256             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 257             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 258
 259             /* Load parameters for j particles */
 260             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 261                                                               charge+jnrC+0,charge+jnrD+0);
 262             vdwjidx0A        = 2*vdwtype[jnrA+0];
 263             vdwjidx0B        = 2*vdwtype[jnrB+0];
 264             vdwjidx0C        = 2*vdwtype[jnrC+0];
 265             vdwjidx0D        = 2*vdwtype[jnrD+0];
 266
 267             fjx0             = _mm_setzero_ps();
 268             fjy0             = _mm_setzero_ps();
 269             fjz0             = _mm_setzero_ps();
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             if (gmx_mm_any_lt(rsq00,rcutoff2))
 276             {
 277
 278             r00              = _mm_mul_ps(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 282                                          vdwparam+vdwioffset0+vdwjidx0B,
 283                                          vdwparam+vdwioffset0+vdwjidx0C,
 284                                          vdwparam+vdwioffset0+vdwjidx0D,
 285                                          &c6_00,&c12_00);
 286             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 287                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 288                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 289                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 290
 291             /* Analytical LJ-PME */
 292             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 293             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 294             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 295             exponent         = sse2_exp_f(ewcljrsq);
 296             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 297             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 298             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 299             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 300             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 301             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) ,
 302                                           _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));
 303             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 304             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);
 305
 306             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 310             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 311
 312             fscal            = fvdw;
 313
 314             fscal            = _mm_and_ps(fscal,cutoff_mask);
 315
 316             /* Calculate temporary vectorial force */
 317             tx               = _mm_mul_ps(fscal,dx00);
 318             ty               = _mm_mul_ps(fscal,dy00);
 319             tz               = _mm_mul_ps(fscal,dz00);
 320
 321             /* Update vectorial force */
 322             fix0             = _mm_add_ps(fix0,tx);
 323             fiy0             = _mm_add_ps(fiy0,ty);
 324             fiz0             = _mm_add_ps(fiz0,tz);
 325
 326             fjx0             = _mm_add_ps(fjx0,tx);
 327             fjy0             = _mm_add_ps(fjy0,ty);
 328             fjz0             = _mm_add_ps(fjz0,tz);
 329
 330             }
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             if (gmx_mm_any_lt(rsq10,rcutoff2))
 337             {
 338
 339             r10              = _mm_mul_ps(rsq10,rinv10);
 340
 341             /* Compute parameters for interactions between i and j atoms */
 342             qq10             = _mm_mul_ps(iq1,jq0);
 343
 344             /* EWALD ELECTROSTATICS */
 345
 346             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 347             ewrt             = _mm_mul_ps(r10,ewtabscale);
 348             ewitab           = _mm_cvttps_epi32(ewrt);
 349             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 350             ewitab           = _mm_slli_epi32(ewitab,2);
 351             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 352             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 353             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 354             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 355             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 356             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 357             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 358             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 359             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 360
 361             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velec            = _mm_and_ps(velec,cutoff_mask);
 365             velecsum         = _mm_add_ps(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             fscal            = _mm_and_ps(fscal,cutoff_mask);
 370
 371             /* Calculate temporary vectorial force */
 372             tx               = _mm_mul_ps(fscal,dx10);
 373             ty               = _mm_mul_ps(fscal,dy10);
 374             tz               = _mm_mul_ps(fscal,dz10);
 375
 376             /* Update vectorial force */
 377             fix1             = _mm_add_ps(fix1,tx);
 378             fiy1             = _mm_add_ps(fiy1,ty);
 379             fiz1             = _mm_add_ps(fiz1,tz);
 380
 381             fjx0             = _mm_add_ps(fjx0,tx);
 382             fjy0             = _mm_add_ps(fjy0,ty);
 383             fjz0             = _mm_add_ps(fjz0,tz);
 384
 385             }
 386
 387             /**************************
 388              * CALCULATE INTERACTIONS *
 389              **************************/
 390
 391             if (gmx_mm_any_lt(rsq20,rcutoff2))
 392             {
 393
 394             r20              = _mm_mul_ps(rsq20,rinv20);
 395
 396             /* Compute parameters for interactions between i and j atoms */
 397             qq20             = _mm_mul_ps(iq2,jq0);
 398
 399             /* EWALD ELECTROSTATICS */
 400
 401             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 402             ewrt             = _mm_mul_ps(r20,ewtabscale);
 403             ewitab           = _mm_cvttps_epi32(ewrt);
 404             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 405             ewitab           = _mm_slli_epi32(ewitab,2);
 406             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 407             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 408             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 409             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 410             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 411             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 412             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 413             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 414             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 415
 416             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velec            = _mm_and_ps(velec,cutoff_mask);
 420             velecsum         = _mm_add_ps(velecsum,velec);
 421
 422             fscal            = felec;
 423
 424             fscal            = _mm_and_ps(fscal,cutoff_mask);
 425
 426             /* Calculate temporary vectorial force */
 427             tx               = _mm_mul_ps(fscal,dx20);
 428             ty               = _mm_mul_ps(fscal,dy20);
 429             tz               = _mm_mul_ps(fscal,dz20);
 430
 431             /* Update vectorial force */
 432             fix2             = _mm_add_ps(fix2,tx);
 433             fiy2             = _mm_add_ps(fiy2,ty);
 434             fiz2             = _mm_add_ps(fiz2,tz);
 435
 436             fjx0             = _mm_add_ps(fjx0,tx);
 437             fjy0             = _mm_add_ps(fjy0,ty);
 438             fjz0             = _mm_add_ps(fjz0,tz);
 439
 440             }
 441
 442             /**************************
 443              * CALCULATE INTERACTIONS *
 444              **************************/
 445
 446             if (gmx_mm_any_lt(rsq30,rcutoff2))
 447             {
 448
 449             r30              = _mm_mul_ps(rsq30,rinv30);
 450
 451             /* Compute parameters for interactions between i and j atoms */
 452             qq30             = _mm_mul_ps(iq3,jq0);
 453
 454             /* EWALD ELECTROSTATICS */
 455
 456             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 457             ewrt             = _mm_mul_ps(r30,ewtabscale);
 458             ewitab           = _mm_cvttps_epi32(ewrt);
 459             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 460             ewitab           = _mm_slli_epi32(ewitab,2);
 461             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 462             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 463             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 464             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 465             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 466             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 467             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 468             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 469             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 470
 471             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 472
 473             /* Update potential sum for this i atom from the interaction with this j atom. */
 474             velec            = _mm_and_ps(velec,cutoff_mask);
 475             velecsum         = _mm_add_ps(velecsum,velec);
 476
 477             fscal            = felec;
 478
 479             fscal            = _mm_and_ps(fscal,cutoff_mask);
 480
 481             /* Calculate temporary vectorial force */
 482             tx               = _mm_mul_ps(fscal,dx30);
 483             ty               = _mm_mul_ps(fscal,dy30);
 484             tz               = _mm_mul_ps(fscal,dz30);
 485
 486             /* Update vectorial force */
 487             fix3             = _mm_add_ps(fix3,tx);
 488             fiy3             = _mm_add_ps(fiy3,ty);
 489             fiz3             = _mm_add_ps(fiz3,tz);
 490
 491             fjx0             = _mm_add_ps(fjx0,tx);
 492             fjy0             = _mm_add_ps(fjy0,ty);
 493             fjz0             = _mm_add_ps(fjz0,tz);
 494
 495             }
 496
 497             fjptrA             = f+j_coord_offsetA;
 498             fjptrB             = f+j_coord_offsetB;
 499             fjptrC             = f+j_coord_offsetC;
 500             fjptrD             = f+j_coord_offsetD;
 501
 502             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 503
 504             /* Inner loop uses 200 flops */
 505         }
 506
 507         if(jidx<j_index_end)
 508         {
 509
 510             /* Get j neighbor index, and coordinate index */
 511             jnrlistA         = jjnr[jidx];
 512             jnrlistB         = jjnr[jidx+1];
 513             jnrlistC         = jjnr[jidx+2];
 514             jnrlistD         = jjnr[jidx+3];
 515             /* Sign of each element will be negative for non-real atoms.
 516              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 517              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 518              */
 519             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 520             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 521             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 522             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 523             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 524             j_coord_offsetA  = DIM*jnrA;
 525             j_coord_offsetB  = DIM*jnrB;
 526             j_coord_offsetC  = DIM*jnrC;
 527             j_coord_offsetD  = DIM*jnrD;
 528
 529             /* load j atom coordinates */
 530             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 531                                               x+j_coord_offsetC,x+j_coord_offsetD,
 532                                               &jx0,&jy0,&jz0);
 533
 534             /* Calculate displacement vector */
 535             dx00             = _mm_sub_ps(ix0,jx0);
 536             dy00             = _mm_sub_ps(iy0,jy0);
 537             dz00             = _mm_sub_ps(iz0,jz0);
 538             dx10             = _mm_sub_ps(ix1,jx0);
 539             dy10             = _mm_sub_ps(iy1,jy0);
 540             dz10             = _mm_sub_ps(iz1,jz0);
 541             dx20             = _mm_sub_ps(ix2,jx0);
 542             dy20             = _mm_sub_ps(iy2,jy0);
 543             dz20             = _mm_sub_ps(iz2,jz0);
 544             dx30             = _mm_sub_ps(ix3,jx0);
 545             dy30             = _mm_sub_ps(iy3,jy0);
 546             dz30             = _mm_sub_ps(iz3,jz0);
 547
 548             /* Calculate squared distance and things based on it */
 549             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 550             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 551             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 552             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 553
 554             rinv00           = sse2_invsqrt_f(rsq00);
 555             rinv10           = sse2_invsqrt_f(rsq10);
 556             rinv20           = sse2_invsqrt_f(rsq20);
 557             rinv30           = sse2_invsqrt_f(rsq30);
 558
 559             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 560             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 561             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 562             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 563
 564             /* Load parameters for j particles */
 565             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 566                                                               charge+jnrC+0,charge+jnrD+0);
 567             vdwjidx0A        = 2*vdwtype[jnrA+0];
 568             vdwjidx0B        = 2*vdwtype[jnrB+0];
 569             vdwjidx0C        = 2*vdwtype[jnrC+0];
 570             vdwjidx0D        = 2*vdwtype[jnrD+0];
 571
 572             fjx0             = _mm_setzero_ps();
 573             fjy0             = _mm_setzero_ps();
 574             fjz0             = _mm_setzero_ps();
 575
 576             /**************************
 577              * CALCULATE INTERACTIONS *
 578              **************************/
 579
 580             if (gmx_mm_any_lt(rsq00,rcutoff2))
 581             {
 582
 583             r00              = _mm_mul_ps(rsq00,rinv00);
 584             r00              = _mm_andnot_ps(dummy_mask,r00);
 585
 586             /* Compute parameters for interactions between i and j atoms */
 587             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 588                                          vdwparam+vdwioffset0+vdwjidx0B,
 589                                          vdwparam+vdwioffset0+vdwjidx0C,
 590                                          vdwparam+vdwioffset0+vdwjidx0D,
 591                                          &c6_00,&c12_00);
 592             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 593                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 594                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 595                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 596
 597             /* Analytical LJ-PME */
 598             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 599             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 600             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 601             exponent         = sse2_exp_f(ewcljrsq);
 602             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 603             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 604             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 605             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 606             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 607             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) ,
 608                                           _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));
 609             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 610             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);
 611
 612             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 616             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 617             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 618
 619             fscal            = fvdw;
 620
 621             fscal            = _mm_and_ps(fscal,cutoff_mask);
 622
 623             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 624
 625             /* Calculate temporary vectorial force */
 626             tx               = _mm_mul_ps(fscal,dx00);
 627             ty               = _mm_mul_ps(fscal,dy00);
 628             tz               = _mm_mul_ps(fscal,dz00);
 629
 630             /* Update vectorial force */
 631             fix0             = _mm_add_ps(fix0,tx);
 632             fiy0             = _mm_add_ps(fiy0,ty);
 633             fiz0             = _mm_add_ps(fiz0,tz);
 634
 635             fjx0             = _mm_add_ps(fjx0,tx);
 636             fjy0             = _mm_add_ps(fjy0,ty);
 637             fjz0             = _mm_add_ps(fjz0,tz);
 638
 639             }
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             if (gmx_mm_any_lt(rsq10,rcutoff2))
 646             {
 647
 648             r10              = _mm_mul_ps(rsq10,rinv10);
 649             r10              = _mm_andnot_ps(dummy_mask,r10);
 650
 651             /* Compute parameters for interactions between i and j atoms */
 652             qq10             = _mm_mul_ps(iq1,jq0);
 653
 654             /* EWALD ELECTROSTATICS */
 655
 656             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 657             ewrt             = _mm_mul_ps(r10,ewtabscale);
 658             ewitab           = _mm_cvttps_epi32(ewrt);
 659             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 660             ewitab           = _mm_slli_epi32(ewitab,2);
 661             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 662             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 663             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 664             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 665             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 666             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 667             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 668             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 669             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 670
 671             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 672
 673             /* Update potential sum for this i atom from the interaction with this j atom. */
 674             velec            = _mm_and_ps(velec,cutoff_mask);
 675             velec            = _mm_andnot_ps(dummy_mask,velec);
 676             velecsum         = _mm_add_ps(velecsum,velec);
 677
 678             fscal            = felec;
 679
 680             fscal            = _mm_and_ps(fscal,cutoff_mask);
 681
 682             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 683
 684             /* Calculate temporary vectorial force */
 685             tx               = _mm_mul_ps(fscal,dx10);
 686             ty               = _mm_mul_ps(fscal,dy10);
 687             tz               = _mm_mul_ps(fscal,dz10);
 688
 689             /* Update vectorial force */
 690             fix1             = _mm_add_ps(fix1,tx);
 691             fiy1             = _mm_add_ps(fiy1,ty);
 692             fiz1             = _mm_add_ps(fiz1,tz);
 693
 694             fjx0             = _mm_add_ps(fjx0,tx);
 695             fjy0             = _mm_add_ps(fjy0,ty);
 696             fjz0             = _mm_add_ps(fjz0,tz);
 697
 698             }
 699
 700             /**************************
 701              * CALCULATE INTERACTIONS *
 702              **************************/
 703
 704             if (gmx_mm_any_lt(rsq20,rcutoff2))
 705             {
 706
 707             r20              = _mm_mul_ps(rsq20,rinv20);
 708             r20              = _mm_andnot_ps(dummy_mask,r20);
 709
 710             /* Compute parameters for interactions between i and j atoms */
 711             qq20             = _mm_mul_ps(iq2,jq0);
 712
 713             /* EWALD ELECTROSTATICS */
 714
 715             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 716             ewrt             = _mm_mul_ps(r20,ewtabscale);
 717             ewitab           = _mm_cvttps_epi32(ewrt);
 718             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 719             ewitab           = _mm_slli_epi32(ewitab,2);
 720             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 721             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 722             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 723             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 724             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 725             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 726             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 727             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 728             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 729
 730             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 731
 732             /* Update potential sum for this i atom from the interaction with this j atom. */
 733             velec            = _mm_and_ps(velec,cutoff_mask);
 734             velec            = _mm_andnot_ps(dummy_mask,velec);
 735             velecsum         = _mm_add_ps(velecsum,velec);
 736
 737             fscal            = felec;
 738
 739             fscal            = _mm_and_ps(fscal,cutoff_mask);
 740
 741             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 742
 743             /* Calculate temporary vectorial force */
 744             tx               = _mm_mul_ps(fscal,dx20);
 745             ty               = _mm_mul_ps(fscal,dy20);
 746             tz               = _mm_mul_ps(fscal,dz20);
 747
 748             /* Update vectorial force */
 749             fix2             = _mm_add_ps(fix2,tx);
 750             fiy2             = _mm_add_ps(fiy2,ty);
 751             fiz2             = _mm_add_ps(fiz2,tz);
 752
 753             fjx0             = _mm_add_ps(fjx0,tx);
 754             fjy0             = _mm_add_ps(fjy0,ty);
 755             fjz0             = _mm_add_ps(fjz0,tz);
 756
 757             }
 758
 759             /**************************
 760              * CALCULATE INTERACTIONS *
 761              **************************/
 762
 763             if (gmx_mm_any_lt(rsq30,rcutoff2))
 764             {
 765
 766             r30              = _mm_mul_ps(rsq30,rinv30);
 767             r30              = _mm_andnot_ps(dummy_mask,r30);
 768
 769             /* Compute parameters for interactions between i and j atoms */
 770             qq30             = _mm_mul_ps(iq3,jq0);
 771
 772             /* EWALD ELECTROSTATICS */
 773
 774             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 775             ewrt             = _mm_mul_ps(r30,ewtabscale);
 776             ewitab           = _mm_cvttps_epi32(ewrt);
 777             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 778             ewitab           = _mm_slli_epi32(ewitab,2);
 779             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 780             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 781             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 782             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 783             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 784             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 785             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 786             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 787             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 788
 789             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 790
 791             /* Update potential sum for this i atom from the interaction with this j atom. */
 792             velec            = _mm_and_ps(velec,cutoff_mask);
 793             velec            = _mm_andnot_ps(dummy_mask,velec);
 794             velecsum         = _mm_add_ps(velecsum,velec);
 795
 796             fscal            = felec;
 797
 798             fscal            = _mm_and_ps(fscal,cutoff_mask);
 799
 800             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 801
 802             /* Calculate temporary vectorial force */
 803             tx               = _mm_mul_ps(fscal,dx30);
 804             ty               = _mm_mul_ps(fscal,dy30);
 805             tz               = _mm_mul_ps(fscal,dz30);
 806
 807             /* Update vectorial force */
 808             fix3             = _mm_add_ps(fix3,tx);
 809             fiy3             = _mm_add_ps(fiy3,ty);
 810             fiz3             = _mm_add_ps(fiz3,tz);
 811
 812             fjx0             = _mm_add_ps(fjx0,tx);
 813             fjy0             = _mm_add_ps(fjy0,ty);
 814             fjz0             = _mm_add_ps(fjz0,tz);
 815
 816             }
 817
 818             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 819             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 820             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 821             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 822
 823             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 824
 825             /* Inner loop uses 204 flops */
 826         }
 827
 828         /* End of innermost loop */
 829
 830         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 831                                               f+i_coord_offset,fshift+i_shift_offset);
 832
 833         ggid                        = gid[iidx];
 834         /* Update potential energies */
 835         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 836         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 837
 838         /* Increment number of inner iterations */
 839         inneriter                  += j_index_end - j_index_start;
 840
 841         /* Outer loop uses 26 flops */
 842     }
 843
 844     /* Increment number of outer iterations */
 845     outeriter        += nri;
 846
 847     /* Update outer/inner flops */
 848
 849     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*204);
 850 }
 851 /*
 852  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_single
 853  * Electrostatics interaction: Ewald
 854  * VdW interaction:            LJEwald
 855  * Geometry:                   Water4-Particle
 856  * Calculate force/pot:        Force
 857  */
 858 void
 859 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse2_single
 860                     (t_nblist                    * gmx_restrict       nlist,
 861                      rvec                        * gmx_restrict          xx,
 862                      rvec                        * gmx_restrict          ff,
 863                      struct t_forcerec           * gmx_restrict          fr,
 864                      t_mdatoms                   * gmx_restrict     mdatoms,
 865                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 866                      t_nrnb                      * gmx_restrict        nrnb)
 867 {
 868     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 869      * just 0 for non-waters.
 870      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 871      * jnr indices corresponding to data put in the four positions in the SIMD register.
 872      */
 873     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 874     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 875     int              jnrA,jnrB,jnrC,jnrD;
 876     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 877     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 878     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 879     real             rcutoff_scalar;
 880     real             *shiftvec,*fshift,*x,*f;
 881     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 882     real             scratch[4*DIM];
 883     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 884     int              vdwioffset0;
 885     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 886     int              vdwioffset1;
 887     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 888     int              vdwioffset2;
 889     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 890     int              vdwioffset3;
 891     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 892     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 893     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 894     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 895     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 896     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 897     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 898     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 899     real             *charge;
 900     int              nvdwtype;
 901     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 902     int              *vdwtype;
 903     real             *vdwparam;
 904     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 905     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 906     __m128           c6grid_00;
 907     __m128           c6grid_10;
 908     __m128           c6grid_20;
 909     __m128           c6grid_30;
 910     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 911     real             *vdwgridparam;
 912     __m128           one_half = _mm_set1_ps(0.5);
 913     __m128           minus_one = _mm_set1_ps(-1.0);
 914     __m128i          ewitab;
 915     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 916     real             *ewtab;
 917     __m128           dummy_mask,cutoff_mask;
 918     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 919     __m128           one     = _mm_set1_ps(1.0);
 920     __m128           two     = _mm_set1_ps(2.0);
 921     x                = xx[0];
 922     f                = ff[0];
 923
 924     nri              = nlist->nri;
 925     iinr             = nlist->iinr;
 926     jindex           = nlist->jindex;
 927     jjnr             = nlist->jjnr;
 928     shiftidx         = nlist->shift;
 929     gid              = nlist->gid;
 930     shiftvec         = fr->shift_vec[0];
 931     fshift           = fr->fshift[0];
 932     facel            = _mm_set1_ps(fr->ic->epsfac);
 933     charge           = mdatoms->chargeA;
 934     nvdwtype         = fr->ntype;
 935     vdwparam         = fr->nbfp;
 936     vdwtype          = mdatoms->typeA;
 937     vdwgridparam     = fr->ljpme_c6grid;
 938     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 939     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 940     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 941
 942     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 943     ewtab            = fr->ic->tabq_coul_F;
 944     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 945     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 946
 947     /* Setup water-specific parameters */
 948     inr              = nlist->iinr[0];
 949     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 950     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 951     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 952     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 953
 954     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 955     rcutoff_scalar   = fr->ic->rcoulomb;
 956     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 957     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 958
 959     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 960     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 961
 962     /* Avoid stupid compiler warnings */
 963     jnrA = jnrB = jnrC = jnrD = 0;
 964     j_coord_offsetA = 0;
 965     j_coord_offsetB = 0;
 966     j_coord_offsetC = 0;
 967     j_coord_offsetD = 0;
 968
 969     outeriter        = 0;
 970     inneriter        = 0;
 971
 972     for(iidx=0;iidx<4*DIM;iidx++)
 973     {
 974         scratch[iidx] = 0.0;
 975     }
 976
 977     /* Start outer loop over neighborlists */
 978     for(iidx=0; iidx<nri; iidx++)
 979     {
 980         /* Load shift vector for this list */
 981         i_shift_offset   = DIM*shiftidx[iidx];
 982
 983         /* Load limits for loop over neighbors */
 984         j_index_start    = jindex[iidx];
 985         j_index_end      = jindex[iidx+1];
 986
 987         /* Get outer coordinate index */
 988         inr              = iinr[iidx];
 989         i_coord_offset   = DIM*inr;
 990
 991         /* Load i particle coords and add shift vector */
 992         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 993                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 994
 995         fix0             = _mm_setzero_ps();
 996         fiy0             = _mm_setzero_ps();
 997         fiz0             = _mm_setzero_ps();
 998         fix1             = _mm_setzero_ps();
 999         fiy1             = _mm_setzero_ps();
1000         fiz1             = _mm_setzero_ps();
1001         fix2             = _mm_setzero_ps();
1002         fiy2             = _mm_setzero_ps();
1003         fiz2             = _mm_setzero_ps();
1004         fix3             = _mm_setzero_ps();
1005         fiy3             = _mm_setzero_ps();
1006         fiz3             = _mm_setzero_ps();
1007
1008         /* Start inner kernel loop */
1009         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1010         {
1011
1012             /* Get j neighbor index, and coordinate index */
1013             jnrA             = jjnr[jidx];
1014             jnrB             = jjnr[jidx+1];
1015             jnrC             = jjnr[jidx+2];
1016             jnrD             = jjnr[jidx+3];
1017             j_coord_offsetA  = DIM*jnrA;
1018             j_coord_offsetB  = DIM*jnrB;
1019             j_coord_offsetC  = DIM*jnrC;
1020             j_coord_offsetD  = DIM*jnrD;
1021
1022             /* load j atom coordinates */
1023             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1024                                               x+j_coord_offsetC,x+j_coord_offsetD,
1025                                               &jx0,&jy0,&jz0);
1026
1027             /* Calculate displacement vector */
1028             dx00             = _mm_sub_ps(ix0,jx0);
1029             dy00             = _mm_sub_ps(iy0,jy0);
1030             dz00             = _mm_sub_ps(iz0,jz0);
1031             dx10             = _mm_sub_ps(ix1,jx0);
1032             dy10             = _mm_sub_ps(iy1,jy0);
1033             dz10             = _mm_sub_ps(iz1,jz0);
1034             dx20             = _mm_sub_ps(ix2,jx0);
1035             dy20             = _mm_sub_ps(iy2,jy0);
1036             dz20             = _mm_sub_ps(iz2,jz0);
1037             dx30             = _mm_sub_ps(ix3,jx0);
1038             dy30             = _mm_sub_ps(iy3,jy0);
1039             dz30             = _mm_sub_ps(iz3,jz0);
1040
1041             /* Calculate squared distance and things based on it */
1042             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1043             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1044             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1045             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1046
1047             rinv00           = sse2_invsqrt_f(rsq00);
1048             rinv10           = sse2_invsqrt_f(rsq10);
1049             rinv20           = sse2_invsqrt_f(rsq20);
1050             rinv30           = sse2_invsqrt_f(rsq30);
1051
1052             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1053             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1054             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1055             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1056
1057             /* Load parameters for j particles */
1058             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1059                                                               charge+jnrC+0,charge+jnrD+0);
1060             vdwjidx0A        = 2*vdwtype[jnrA+0];
1061             vdwjidx0B        = 2*vdwtype[jnrB+0];
1062             vdwjidx0C        = 2*vdwtype[jnrC+0];
1063             vdwjidx0D        = 2*vdwtype[jnrD+0];
1064
1065             fjx0             = _mm_setzero_ps();
1066             fjy0             = _mm_setzero_ps();
1067             fjz0             = _mm_setzero_ps();
1068
1069             /**************************
1070              * CALCULATE INTERACTIONS *
1071              **************************/
1072
1073             if (gmx_mm_any_lt(rsq00,rcutoff2))
1074             {
1075
1076             r00              = _mm_mul_ps(rsq00,rinv00);
1077
1078             /* Compute parameters for interactions between i and j atoms */
1079             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1080                                          vdwparam+vdwioffset0+vdwjidx0B,
1081                                          vdwparam+vdwioffset0+vdwjidx0C,
1082                                          vdwparam+vdwioffset0+vdwjidx0D,
1083                                          &c6_00,&c12_00);
1084             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1085                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1086                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1087                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1088
1089             /* Analytical LJ-PME */
1090             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1091             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1092             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1093             exponent         = sse2_exp_f(ewcljrsq);
1094             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1095             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1096             /* f6A = 6 * C6grid * (1 - poly) */
1097             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1098             /* f6B = C6grid * exponent * beta^6 */
1099             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1100             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1101             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);
1102
1103             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1104
1105             fscal            = fvdw;
1106
1107             fscal            = _mm_and_ps(fscal,cutoff_mask);
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm_mul_ps(fscal,dx00);
1111             ty               = _mm_mul_ps(fscal,dy00);
1112             tz               = _mm_mul_ps(fscal,dz00);
1113
1114             /* Update vectorial force */
1115             fix0             = _mm_add_ps(fix0,tx);
1116             fiy0             = _mm_add_ps(fiy0,ty);
1117             fiz0             = _mm_add_ps(fiz0,tz);
1118
1119             fjx0             = _mm_add_ps(fjx0,tx);
1120             fjy0             = _mm_add_ps(fjy0,ty);
1121             fjz0             = _mm_add_ps(fjz0,tz);
1122
1123             }
1124
1125             /**************************
1126              * CALCULATE INTERACTIONS *
1127              **************************/
1128
1129             if (gmx_mm_any_lt(rsq10,rcutoff2))
1130             {
1131
1132             r10              = _mm_mul_ps(rsq10,rinv10);
1133
1134             /* Compute parameters for interactions between i and j atoms */
1135             qq10             = _mm_mul_ps(iq1,jq0);
1136
1137             /* EWALD ELECTROSTATICS */
1138
1139             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1140             ewrt             = _mm_mul_ps(r10,ewtabscale);
1141             ewitab           = _mm_cvttps_epi32(ewrt);
1142             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1143             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1144                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1145                                          &ewtabF,&ewtabFn);
1146             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1147             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1148
1149             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1150
1151             fscal            = felec;
1152
1153             fscal            = _mm_and_ps(fscal,cutoff_mask);
1154
1155             /* Calculate temporary vectorial force */
1156             tx               = _mm_mul_ps(fscal,dx10);
1157             ty               = _mm_mul_ps(fscal,dy10);
1158             tz               = _mm_mul_ps(fscal,dz10);
1159
1160             /* Update vectorial force */
1161             fix1             = _mm_add_ps(fix1,tx);
1162             fiy1             = _mm_add_ps(fiy1,ty);
1163             fiz1             = _mm_add_ps(fiz1,tz);
1164
1165             fjx0             = _mm_add_ps(fjx0,tx);
1166             fjy0             = _mm_add_ps(fjy0,ty);
1167             fjz0             = _mm_add_ps(fjz0,tz);
1168
1169             }
1170
1171             /**************************
1172              * CALCULATE INTERACTIONS *
1173              **************************/
1174
1175             if (gmx_mm_any_lt(rsq20,rcutoff2))
1176             {
1177
1178             r20              = _mm_mul_ps(rsq20,rinv20);
1179
1180             /* Compute parameters for interactions between i and j atoms */
1181             qq20             = _mm_mul_ps(iq2,jq0);
1182
1183             /* EWALD ELECTROSTATICS */
1184
1185             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1186             ewrt             = _mm_mul_ps(r20,ewtabscale);
1187             ewitab           = _mm_cvttps_epi32(ewrt);
1188             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1189             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1190                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1191                                          &ewtabF,&ewtabFn);
1192             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1193             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1194
1195             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1196
1197             fscal            = felec;
1198
1199             fscal            = _mm_and_ps(fscal,cutoff_mask);
1200
1201             /* Calculate temporary vectorial force */
1202             tx               = _mm_mul_ps(fscal,dx20);
1203             ty               = _mm_mul_ps(fscal,dy20);
1204             tz               = _mm_mul_ps(fscal,dz20);
1205
1206             /* Update vectorial force */
1207             fix2             = _mm_add_ps(fix2,tx);
1208             fiy2             = _mm_add_ps(fiy2,ty);
1209             fiz2             = _mm_add_ps(fiz2,tz);
1210
1211             fjx0             = _mm_add_ps(fjx0,tx);
1212             fjy0             = _mm_add_ps(fjy0,ty);
1213             fjz0             = _mm_add_ps(fjz0,tz);
1214
1215             }
1216
1217             /**************************
1218              * CALCULATE INTERACTIONS *
1219              **************************/
1220
1221             if (gmx_mm_any_lt(rsq30,rcutoff2))
1222             {
1223
1224             r30              = _mm_mul_ps(rsq30,rinv30);
1225
1226             /* Compute parameters for interactions between i and j atoms */
1227             qq30             = _mm_mul_ps(iq3,jq0);
1228
1229             /* EWALD ELECTROSTATICS */
1230
1231             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1232             ewrt             = _mm_mul_ps(r30,ewtabscale);
1233             ewitab           = _mm_cvttps_epi32(ewrt);
1234             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1235             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1236                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1237                                          &ewtabF,&ewtabFn);
1238             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1239             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1240
1241             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1242
1243             fscal            = felec;
1244
1245             fscal            = _mm_and_ps(fscal,cutoff_mask);
1246
1247             /* Calculate temporary vectorial force */
1248             tx               = _mm_mul_ps(fscal,dx30);
1249             ty               = _mm_mul_ps(fscal,dy30);
1250             tz               = _mm_mul_ps(fscal,dz30);
1251
1252             /* Update vectorial force */
1253             fix3             = _mm_add_ps(fix3,tx);
1254             fiy3             = _mm_add_ps(fiy3,ty);
1255             fiz3             = _mm_add_ps(fiz3,tz);
1256
1257             fjx0             = _mm_add_ps(fjx0,tx);
1258             fjy0             = _mm_add_ps(fjy0,ty);
1259             fjz0             = _mm_add_ps(fjz0,tz);
1260
1261             }
1262
1263             fjptrA             = f+j_coord_offsetA;
1264             fjptrB             = f+j_coord_offsetB;
1265             fjptrC             = f+j_coord_offsetC;
1266             fjptrD             = f+j_coord_offsetD;
1267
1268             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1269
1270             /* Inner loop uses 166 flops */
1271         }
1272
1273         if(jidx<j_index_end)
1274         {
1275
1276             /* Get j neighbor index, and coordinate index */
1277             jnrlistA         = jjnr[jidx];
1278             jnrlistB         = jjnr[jidx+1];
1279             jnrlistC         = jjnr[jidx+2];
1280             jnrlistD         = jjnr[jidx+3];
1281             /* Sign of each element will be negative for non-real atoms.
1282              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1283              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1284              */
1285             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1286             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1287             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1288             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1289             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1290             j_coord_offsetA  = DIM*jnrA;
1291             j_coord_offsetB  = DIM*jnrB;
1292             j_coord_offsetC  = DIM*jnrC;
1293             j_coord_offsetD  = DIM*jnrD;
1294
1295             /* load j atom coordinates */
1296             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1297                                               x+j_coord_offsetC,x+j_coord_offsetD,
1298                                               &jx0,&jy0,&jz0);
1299
1300             /* Calculate displacement vector */
1301             dx00             = _mm_sub_ps(ix0,jx0);
1302             dy00             = _mm_sub_ps(iy0,jy0);
1303             dz00             = _mm_sub_ps(iz0,jz0);
1304             dx10             = _mm_sub_ps(ix1,jx0);
1305             dy10             = _mm_sub_ps(iy1,jy0);
1306             dz10             = _mm_sub_ps(iz1,jz0);
1307             dx20             = _mm_sub_ps(ix2,jx0);
1308             dy20             = _mm_sub_ps(iy2,jy0);
1309             dz20             = _mm_sub_ps(iz2,jz0);
1310             dx30             = _mm_sub_ps(ix3,jx0);
1311             dy30             = _mm_sub_ps(iy3,jy0);
1312             dz30             = _mm_sub_ps(iz3,jz0);
1313
1314             /* Calculate squared distance and things based on it */
1315             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1316             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1317             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1318             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1319
1320             rinv00           = sse2_invsqrt_f(rsq00);
1321             rinv10           = sse2_invsqrt_f(rsq10);
1322             rinv20           = sse2_invsqrt_f(rsq20);
1323             rinv30           = sse2_invsqrt_f(rsq30);
1324
1325             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1326             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1327             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1328             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1329
1330             /* Load parameters for j particles */
1331             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1332                                                               charge+jnrC+0,charge+jnrD+0);
1333             vdwjidx0A        = 2*vdwtype[jnrA+0];
1334             vdwjidx0B        = 2*vdwtype[jnrB+0];
1335             vdwjidx0C        = 2*vdwtype[jnrC+0];
1336             vdwjidx0D        = 2*vdwtype[jnrD+0];
1337
1338             fjx0             = _mm_setzero_ps();
1339             fjy0             = _mm_setzero_ps();
1340             fjz0             = _mm_setzero_ps();
1341
1342             /**************************
1343              * CALCULATE INTERACTIONS *
1344              **************************/
1345
1346             if (gmx_mm_any_lt(rsq00,rcutoff2))
1347             {
1348
1349             r00              = _mm_mul_ps(rsq00,rinv00);
1350             r00              = _mm_andnot_ps(dummy_mask,r00);
1351
1352             /* Compute parameters for interactions between i and j atoms */
1353             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1354                                          vdwparam+vdwioffset0+vdwjidx0B,
1355                                          vdwparam+vdwioffset0+vdwjidx0C,
1356                                          vdwparam+vdwioffset0+vdwjidx0D,
1357                                          &c6_00,&c12_00);
1358             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1359                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1360                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1361                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1362
1363             /* Analytical LJ-PME */
1364             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1365             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1366             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1367             exponent         = sse2_exp_f(ewcljrsq);
1368             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1369             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1370             /* f6A = 6 * C6grid * (1 - poly) */
1371             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1372             /* f6B = C6grid * exponent * beta^6 */
1373             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1374             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1375             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);
1376
1377             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1378
1379             fscal            = fvdw;
1380
1381             fscal            = _mm_and_ps(fscal,cutoff_mask);
1382
1383             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1384
1385             /* Calculate temporary vectorial force */
1386             tx               = _mm_mul_ps(fscal,dx00);
1387             ty               = _mm_mul_ps(fscal,dy00);
1388             tz               = _mm_mul_ps(fscal,dz00);
1389
1390             /* Update vectorial force */
1391             fix0             = _mm_add_ps(fix0,tx);
1392             fiy0             = _mm_add_ps(fiy0,ty);
1393             fiz0             = _mm_add_ps(fiz0,tz);
1394
1395             fjx0             = _mm_add_ps(fjx0,tx);
1396             fjy0             = _mm_add_ps(fjy0,ty);
1397             fjz0             = _mm_add_ps(fjz0,tz);
1398
1399             }
1400
1401             /**************************
1402              * CALCULATE INTERACTIONS *
1403              **************************/
1404
1405             if (gmx_mm_any_lt(rsq10,rcutoff2))
1406             {
1407
1408             r10              = _mm_mul_ps(rsq10,rinv10);
1409             r10              = _mm_andnot_ps(dummy_mask,r10);
1410
1411             /* Compute parameters for interactions between i and j atoms */
1412             qq10             = _mm_mul_ps(iq1,jq0);
1413
1414             /* EWALD ELECTROSTATICS */
1415
1416             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1417             ewrt             = _mm_mul_ps(r10,ewtabscale);
1418             ewitab           = _mm_cvttps_epi32(ewrt);
1419             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1420             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1421                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1422                                          &ewtabF,&ewtabFn);
1423             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1424             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1425
1426             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1427
1428             fscal            = felec;
1429
1430             fscal            = _mm_and_ps(fscal,cutoff_mask);
1431
1432             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1433
1434             /* Calculate temporary vectorial force */
1435             tx               = _mm_mul_ps(fscal,dx10);
1436             ty               = _mm_mul_ps(fscal,dy10);
1437             tz               = _mm_mul_ps(fscal,dz10);
1438
1439             /* Update vectorial force */
1440             fix1             = _mm_add_ps(fix1,tx);
1441             fiy1             = _mm_add_ps(fiy1,ty);
1442             fiz1             = _mm_add_ps(fiz1,tz);
1443
1444             fjx0             = _mm_add_ps(fjx0,tx);
1445             fjy0             = _mm_add_ps(fjy0,ty);
1446             fjz0             = _mm_add_ps(fjz0,tz);
1447
1448             }
1449
1450             /**************************
1451              * CALCULATE INTERACTIONS *
1452              **************************/
1453
1454             if (gmx_mm_any_lt(rsq20,rcutoff2))
1455             {
1456
1457             r20              = _mm_mul_ps(rsq20,rinv20);
1458             r20              = _mm_andnot_ps(dummy_mask,r20);
1459
1460             /* Compute parameters for interactions between i and j atoms */
1461             qq20             = _mm_mul_ps(iq2,jq0);
1462
1463             /* EWALD ELECTROSTATICS */
1464
1465             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1466             ewrt             = _mm_mul_ps(r20,ewtabscale);
1467             ewitab           = _mm_cvttps_epi32(ewrt);
1468             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1469             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1470                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1471                                          &ewtabF,&ewtabFn);
1472             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1473             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1474
1475             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1476
1477             fscal            = felec;
1478
1479             fscal            = _mm_and_ps(fscal,cutoff_mask);
1480
1481             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1482
1483             /* Calculate temporary vectorial force */
1484             tx               = _mm_mul_ps(fscal,dx20);
1485             ty               = _mm_mul_ps(fscal,dy20);
1486             tz               = _mm_mul_ps(fscal,dz20);
1487
1488             /* Update vectorial force */
1489             fix2             = _mm_add_ps(fix2,tx);
1490             fiy2             = _mm_add_ps(fiy2,ty);
1491             fiz2             = _mm_add_ps(fiz2,tz);
1492
1493             fjx0             = _mm_add_ps(fjx0,tx);
1494             fjy0             = _mm_add_ps(fjy0,ty);
1495             fjz0             = _mm_add_ps(fjz0,tz);
1496
1497             }
1498
1499             /**************************
1500              * CALCULATE INTERACTIONS *
1501              **************************/
1502
1503             if (gmx_mm_any_lt(rsq30,rcutoff2))
1504             {
1505
1506             r30              = _mm_mul_ps(rsq30,rinv30);
1507             r30              = _mm_andnot_ps(dummy_mask,r30);
1508
1509             /* Compute parameters for interactions between i and j atoms */
1510             qq30             = _mm_mul_ps(iq3,jq0);
1511
1512             /* EWALD ELECTROSTATICS */
1513
1514             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1515             ewrt             = _mm_mul_ps(r30,ewtabscale);
1516             ewitab           = _mm_cvttps_epi32(ewrt);
1517             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1518             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1519                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1520                                          &ewtabF,&ewtabFn);
1521             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1522             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1523
1524             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1525
1526             fscal            = felec;
1527
1528             fscal            = _mm_and_ps(fscal,cutoff_mask);
1529
1530             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1531
1532             /* Calculate temporary vectorial force */
1533             tx               = _mm_mul_ps(fscal,dx30);
1534             ty               = _mm_mul_ps(fscal,dy30);
1535             tz               = _mm_mul_ps(fscal,dz30);
1536
1537             /* Update vectorial force */
1538             fix3             = _mm_add_ps(fix3,tx);
1539             fiy3             = _mm_add_ps(fiy3,ty);
1540             fiz3             = _mm_add_ps(fiz3,tz);
1541
1542             fjx0             = _mm_add_ps(fjx0,tx);
1543             fjy0             = _mm_add_ps(fjy0,ty);
1544             fjz0             = _mm_add_ps(fjz0,tz);
1545
1546             }
1547
1548             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1549             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1550             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1551             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1552
1553             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1554
1555             /* Inner loop uses 170 flops */
1556         }
1557
1558         /* End of innermost loop */
1559
1560         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1561                                               f+i_coord_offset,fshift+i_shift_offset);
1562
1563         /* Increment number of inner iterations */
1564         inneriter                  += j_index_end - j_index_start;
1565
1566         /* Outer loop uses 24 flops */
1567     }
1568
1569     /* Increment number of outer iterations */
1570     outeriter        += nri;
1571
1572     /* Update outer/inner flops */
1573
1574     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*170);
1575 }