src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJEw_GeomW4P1_sse4_1_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 sse4_1_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_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse4_1_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse4_1_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     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = jnrC = jnrD = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156     j_coord_offsetC = 0;
 157     j_coord_offsetD = 0;
 158
 159     outeriter        = 0;
 160     inneriter        = 0;
 161
 162     for(iidx=0;iidx<4*DIM;iidx++)
 163     {
 164         scratch[iidx] = 0.0;
 165     }
 166
 167     /* Start outer loop over neighborlists */
 168     for(iidx=0; iidx<nri; iidx++)
 169     {
 170         /* Load shift vector for this list */
 171         i_shift_offset   = DIM*shiftidx[iidx];
 172
 173         /* Load limits for loop over neighbors */
 174         j_index_start    = jindex[iidx];
 175         j_index_end      = jindex[iidx+1];
 176
 177         /* Get outer coordinate index */
 178         inr              = iinr[iidx];
 179         i_coord_offset   = DIM*inr;
 180
 181         /* Load i particle coords and add shift vector */
 182         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 183                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 184
 185         fix0             = _mm_setzero_ps();
 186         fiy0             = _mm_setzero_ps();
 187         fiz0             = _mm_setzero_ps();
 188         fix1             = _mm_setzero_ps();
 189         fiy1             = _mm_setzero_ps();
 190         fiz1             = _mm_setzero_ps();
 191         fix2             = _mm_setzero_ps();
 192         fiy2             = _mm_setzero_ps();
 193         fiz2             = _mm_setzero_ps();
 194         fix3             = _mm_setzero_ps();
 195         fiy3             = _mm_setzero_ps();
 196         fiz3             = _mm_setzero_ps();
 197
 198         /* Reset potential sums */
 199         velecsum         = _mm_setzero_ps();
 200         vvdwsum          = _mm_setzero_ps();
 201
 202         /* Start inner kernel loop */
 203         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 204         {
 205
 206             /* Get j neighbor index, and coordinate index */
 207             jnrA             = jjnr[jidx];
 208             jnrB             = jjnr[jidx+1];
 209             jnrC             = jjnr[jidx+2];
 210             jnrD             = jjnr[jidx+3];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215
 216             /* load j atom coordinates */
 217             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                               x+j_coord_offsetC,x+j_coord_offsetD,
 219                                               &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm_sub_ps(ix0,jx0);
 223             dy00             = _mm_sub_ps(iy0,jy0);
 224             dz00             = _mm_sub_ps(iz0,jz0);
 225             dx10             = _mm_sub_ps(ix1,jx0);
 226             dy10             = _mm_sub_ps(iy1,jy0);
 227             dz10             = _mm_sub_ps(iz1,jz0);
 228             dx20             = _mm_sub_ps(ix2,jx0);
 229             dy20             = _mm_sub_ps(iy2,jy0);
 230             dz20             = _mm_sub_ps(iz2,jz0);
 231             dx30             = _mm_sub_ps(ix3,jx0);
 232             dy30             = _mm_sub_ps(iy3,jy0);
 233             dz30             = _mm_sub_ps(iz3,jz0);
 234
 235             /* Calculate squared distance and things based on it */
 236             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 237             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 238             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 239             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 240
 241             rinv00           = sse41_invsqrt_f(rsq00);
 242             rinv10           = sse41_invsqrt_f(rsq10);
 243             rinv20           = sse41_invsqrt_f(rsq20);
 244             rinv30           = sse41_invsqrt_f(rsq30);
 245
 246             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 247             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 248             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 249             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 250
 251             /* Load parameters for j particles */
 252             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 253                                                               charge+jnrC+0,charge+jnrD+0);
 254             vdwjidx0A        = 2*vdwtype[jnrA+0];
 255             vdwjidx0B        = 2*vdwtype[jnrB+0];
 256             vdwjidx0C        = 2*vdwtype[jnrC+0];
 257             vdwjidx0D        = 2*vdwtype[jnrD+0];
 258
 259             fjx0             = _mm_setzero_ps();
 260             fjy0             = _mm_setzero_ps();
 261             fjz0             = _mm_setzero_ps();
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             r00              = _mm_mul_ps(rsq00,rinv00);
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 271                                          vdwparam+vdwioffset0+vdwjidx0B,
 272                                          vdwparam+vdwioffset0+vdwjidx0C,
 273                                          vdwparam+vdwioffset0+vdwjidx0D,
 274                                          &c6_00,&c12_00);
 275
 276             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 277                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 278                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 279                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 280
 281             /* Analytical LJ-PME */
 282             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 283             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 284             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 285             exponent         = sse41_exp_f(ewcljrsq);
 286             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 287             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 288             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 289             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 290             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 291             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 292             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 293             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);
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 297
 298             fscal            = fvdw;
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm_mul_ps(fscal,dx00);
 302             ty               = _mm_mul_ps(fscal,dy00);
 303             tz               = _mm_mul_ps(fscal,dz00);
 304
 305             /* Update vectorial force */
 306             fix0             = _mm_add_ps(fix0,tx);
 307             fiy0             = _mm_add_ps(fiy0,ty);
 308             fiz0             = _mm_add_ps(fiz0,tz);
 309
 310             fjx0             = _mm_add_ps(fjx0,tx);
 311             fjy0             = _mm_add_ps(fjy0,ty);
 312             fjz0             = _mm_add_ps(fjz0,tz);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r10              = _mm_mul_ps(rsq10,rinv10);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm_mul_ps(iq1,jq0);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 326             ewrt             = _mm_mul_ps(r10,ewtabscale);
 327             ewitab           = _mm_cvttps_epi32(ewrt);
 328             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 329             ewitab           = _mm_slli_epi32(ewitab,2);
 330             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 331             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 332             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 333             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 334             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 335             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 336             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 337             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 338             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velecsum         = _mm_add_ps(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm_mul_ps(fscal,dx10);
 347             ty               = _mm_mul_ps(fscal,dy10);
 348             tz               = _mm_mul_ps(fscal,dz10);
 349
 350             /* Update vectorial force */
 351             fix1             = _mm_add_ps(fix1,tx);
 352             fiy1             = _mm_add_ps(fiy1,ty);
 353             fiz1             = _mm_add_ps(fiz1,tz);
 354
 355             fjx0             = _mm_add_ps(fjx0,tx);
 356             fjy0             = _mm_add_ps(fjy0,ty);
 357             fjz0             = _mm_add_ps(fjz0,tz);
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             r20              = _mm_mul_ps(rsq20,rinv20);
 364
 365             /* Compute parameters for interactions between i and j atoms */
 366             qq20             = _mm_mul_ps(iq2,jq0);
 367
 368             /* EWALD ELECTROSTATICS */
 369
 370             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 371             ewrt             = _mm_mul_ps(r20,ewtabscale);
 372             ewitab           = _mm_cvttps_epi32(ewrt);
 373             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 374             ewitab           = _mm_slli_epi32(ewitab,2);
 375             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 376             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 377             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 378             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 379             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 380             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 381             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 382             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 383             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velecsum         = _mm_add_ps(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390             /* Calculate temporary vectorial force */
 391             tx               = _mm_mul_ps(fscal,dx20);
 392             ty               = _mm_mul_ps(fscal,dy20);
 393             tz               = _mm_mul_ps(fscal,dz20);
 394
 395             /* Update vectorial force */
 396             fix2             = _mm_add_ps(fix2,tx);
 397             fiy2             = _mm_add_ps(fiy2,ty);
 398             fiz2             = _mm_add_ps(fiz2,tz);
 399
 400             fjx0             = _mm_add_ps(fjx0,tx);
 401             fjy0             = _mm_add_ps(fjy0,ty);
 402             fjz0             = _mm_add_ps(fjz0,tz);
 403
 404             /**************************
 405              * CALCULATE INTERACTIONS *
 406              **************************/
 407
 408             r30              = _mm_mul_ps(rsq30,rinv30);
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq30             = _mm_mul_ps(iq3,jq0);
 412
 413             /* EWALD ELECTROSTATICS */
 414
 415             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 416             ewrt             = _mm_mul_ps(r30,ewtabscale);
 417             ewitab           = _mm_cvttps_epi32(ewrt);
 418             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 419             ewitab           = _mm_slli_epi32(ewitab,2);
 420             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 421             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 422             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 423             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 424             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 425             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 426             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 427             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 428             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velecsum         = _mm_add_ps(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435             /* Calculate temporary vectorial force */
 436             tx               = _mm_mul_ps(fscal,dx30);
 437             ty               = _mm_mul_ps(fscal,dy30);
 438             tz               = _mm_mul_ps(fscal,dz30);
 439
 440             /* Update vectorial force */
 441             fix3             = _mm_add_ps(fix3,tx);
 442             fiy3             = _mm_add_ps(fiy3,ty);
 443             fiz3             = _mm_add_ps(fiz3,tz);
 444
 445             fjx0             = _mm_add_ps(fjx0,tx);
 446             fjy0             = _mm_add_ps(fjy0,ty);
 447             fjz0             = _mm_add_ps(fjz0,tz);
 448
 449             fjptrA             = f+j_coord_offsetA;
 450             fjptrB             = f+j_coord_offsetB;
 451             fjptrC             = f+j_coord_offsetC;
 452             fjptrD             = f+j_coord_offsetD;
 453
 454             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 455
 456             /* Inner loop uses 174 flops */
 457         }
 458
 459         if(jidx<j_index_end)
 460         {
 461
 462             /* Get j neighbor index, and coordinate index */
 463             jnrlistA         = jjnr[jidx];
 464             jnrlistB         = jjnr[jidx+1];
 465             jnrlistC         = jjnr[jidx+2];
 466             jnrlistD         = jjnr[jidx+3];
 467             /* Sign of each element will be negative for non-real atoms.
 468              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 469              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 470              */
 471             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 472             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 473             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 474             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 475             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 476             j_coord_offsetA  = DIM*jnrA;
 477             j_coord_offsetB  = DIM*jnrB;
 478             j_coord_offsetC  = DIM*jnrC;
 479             j_coord_offsetD  = DIM*jnrD;
 480
 481             /* load j atom coordinates */
 482             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 483                                               x+j_coord_offsetC,x+j_coord_offsetD,
 484                                               &jx0,&jy0,&jz0);
 485
 486             /* Calculate displacement vector */
 487             dx00             = _mm_sub_ps(ix0,jx0);
 488             dy00             = _mm_sub_ps(iy0,jy0);
 489             dz00             = _mm_sub_ps(iz0,jz0);
 490             dx10             = _mm_sub_ps(ix1,jx0);
 491             dy10             = _mm_sub_ps(iy1,jy0);
 492             dz10             = _mm_sub_ps(iz1,jz0);
 493             dx20             = _mm_sub_ps(ix2,jx0);
 494             dy20             = _mm_sub_ps(iy2,jy0);
 495             dz20             = _mm_sub_ps(iz2,jz0);
 496             dx30             = _mm_sub_ps(ix3,jx0);
 497             dy30             = _mm_sub_ps(iy3,jy0);
 498             dz30             = _mm_sub_ps(iz3,jz0);
 499
 500             /* Calculate squared distance and things based on it */
 501             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 502             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 503             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 504             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 505
 506             rinv00           = sse41_invsqrt_f(rsq00);
 507             rinv10           = sse41_invsqrt_f(rsq10);
 508             rinv20           = sse41_invsqrt_f(rsq20);
 509             rinv30           = sse41_invsqrt_f(rsq30);
 510
 511             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 512             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 513             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 514             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 515
 516             /* Load parameters for j particles */
 517             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 518                                                               charge+jnrC+0,charge+jnrD+0);
 519             vdwjidx0A        = 2*vdwtype[jnrA+0];
 520             vdwjidx0B        = 2*vdwtype[jnrB+0];
 521             vdwjidx0C        = 2*vdwtype[jnrC+0];
 522             vdwjidx0D        = 2*vdwtype[jnrD+0];
 523
 524             fjx0             = _mm_setzero_ps();
 525             fjy0             = _mm_setzero_ps();
 526             fjz0             = _mm_setzero_ps();
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             r00              = _mm_mul_ps(rsq00,rinv00);
 533             r00              = _mm_andnot_ps(dummy_mask,r00);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 537                                          vdwparam+vdwioffset0+vdwjidx0B,
 538                                          vdwparam+vdwioffset0+vdwjidx0C,
 539                                          vdwparam+vdwioffset0+vdwjidx0D,
 540                                          &c6_00,&c12_00);
 541
 542             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 543                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 544                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 545                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 546
 547             /* Analytical LJ-PME */
 548             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 549             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 550             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 551             exponent         = sse41_exp_f(ewcljrsq);
 552             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 553             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 554             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 555             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 556             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 557             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 558             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 559             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);
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 563             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 564
 565             fscal            = fvdw;
 566
 567             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 568
 569             /* Calculate temporary vectorial force */
 570             tx               = _mm_mul_ps(fscal,dx00);
 571             ty               = _mm_mul_ps(fscal,dy00);
 572             tz               = _mm_mul_ps(fscal,dz00);
 573
 574             /* Update vectorial force */
 575             fix0             = _mm_add_ps(fix0,tx);
 576             fiy0             = _mm_add_ps(fiy0,ty);
 577             fiz0             = _mm_add_ps(fiz0,tz);
 578
 579             fjx0             = _mm_add_ps(fjx0,tx);
 580             fjy0             = _mm_add_ps(fjy0,ty);
 581             fjz0             = _mm_add_ps(fjz0,tz);
 582
 583             /**************************
 584              * CALCULATE INTERACTIONS *
 585              **************************/
 586
 587             r10              = _mm_mul_ps(rsq10,rinv10);
 588             r10              = _mm_andnot_ps(dummy_mask,r10);
 589
 590             /* Compute parameters for interactions between i and j atoms */
 591             qq10             = _mm_mul_ps(iq1,jq0);
 592
 593             /* EWALD ELECTROSTATICS */
 594
 595             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 596             ewrt             = _mm_mul_ps(r10,ewtabscale);
 597             ewitab           = _mm_cvttps_epi32(ewrt);
 598             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 599             ewitab           = _mm_slli_epi32(ewitab,2);
 600             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 601             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 602             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 603             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 604             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 605             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 606             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 607             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 608             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 609
 610             /* Update potential sum for this i atom from the interaction with this j atom. */
 611             velec            = _mm_andnot_ps(dummy_mask,velec);
 612             velecsum         = _mm_add_ps(velecsum,velec);
 613
 614             fscal            = felec;
 615
 616             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 617
 618             /* Calculate temporary vectorial force */
 619             tx               = _mm_mul_ps(fscal,dx10);
 620             ty               = _mm_mul_ps(fscal,dy10);
 621             tz               = _mm_mul_ps(fscal,dz10);
 622
 623             /* Update vectorial force */
 624             fix1             = _mm_add_ps(fix1,tx);
 625             fiy1             = _mm_add_ps(fiy1,ty);
 626             fiz1             = _mm_add_ps(fiz1,tz);
 627
 628             fjx0             = _mm_add_ps(fjx0,tx);
 629             fjy0             = _mm_add_ps(fjy0,ty);
 630             fjz0             = _mm_add_ps(fjz0,tz);
 631
 632             /**************************
 633              * CALCULATE INTERACTIONS *
 634              **************************/
 635
 636             r20              = _mm_mul_ps(rsq20,rinv20);
 637             r20              = _mm_andnot_ps(dummy_mask,r20);
 638
 639             /* Compute parameters for interactions between i and j atoms */
 640             qq20             = _mm_mul_ps(iq2,jq0);
 641
 642             /* EWALD ELECTROSTATICS */
 643
 644             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 645             ewrt             = _mm_mul_ps(r20,ewtabscale);
 646             ewitab           = _mm_cvttps_epi32(ewrt);
 647             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 648             ewitab           = _mm_slli_epi32(ewitab,2);
 649             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 650             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 651             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 652             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 653             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 654             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 655             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 656             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 657             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 658
 659             /* Update potential sum for this i atom from the interaction with this j atom. */
 660             velec            = _mm_andnot_ps(dummy_mask,velec);
 661             velecsum         = _mm_add_ps(velecsum,velec);
 662
 663             fscal            = felec;
 664
 665             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 666
 667             /* Calculate temporary vectorial force */
 668             tx               = _mm_mul_ps(fscal,dx20);
 669             ty               = _mm_mul_ps(fscal,dy20);
 670             tz               = _mm_mul_ps(fscal,dz20);
 671
 672             /* Update vectorial force */
 673             fix2             = _mm_add_ps(fix2,tx);
 674             fiy2             = _mm_add_ps(fiy2,ty);
 675             fiz2             = _mm_add_ps(fiz2,tz);
 676
 677             fjx0             = _mm_add_ps(fjx0,tx);
 678             fjy0             = _mm_add_ps(fjy0,ty);
 679             fjz0             = _mm_add_ps(fjz0,tz);
 680
 681             /**************************
 682              * CALCULATE INTERACTIONS *
 683              **************************/
 684
 685             r30              = _mm_mul_ps(rsq30,rinv30);
 686             r30              = _mm_andnot_ps(dummy_mask,r30);
 687
 688             /* Compute parameters for interactions between i and j atoms */
 689             qq30             = _mm_mul_ps(iq3,jq0);
 690
 691             /* EWALD ELECTROSTATICS */
 692
 693             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 694             ewrt             = _mm_mul_ps(r30,ewtabscale);
 695             ewitab           = _mm_cvttps_epi32(ewrt);
 696             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 697             ewitab           = _mm_slli_epi32(ewitab,2);
 698             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 699             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 700             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 701             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 702             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 703             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 704             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 705             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 706             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 707
 708             /* Update potential sum for this i atom from the interaction with this j atom. */
 709             velec            = _mm_andnot_ps(dummy_mask,velec);
 710             velecsum         = _mm_add_ps(velecsum,velec);
 711
 712             fscal            = felec;
 713
 714             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 715
 716             /* Calculate temporary vectorial force */
 717             tx               = _mm_mul_ps(fscal,dx30);
 718             ty               = _mm_mul_ps(fscal,dy30);
 719             tz               = _mm_mul_ps(fscal,dz30);
 720
 721             /* Update vectorial force */
 722             fix3             = _mm_add_ps(fix3,tx);
 723             fiy3             = _mm_add_ps(fiy3,ty);
 724             fiz3             = _mm_add_ps(fiz3,tz);
 725
 726             fjx0             = _mm_add_ps(fjx0,tx);
 727             fjy0             = _mm_add_ps(fjy0,ty);
 728             fjz0             = _mm_add_ps(fjz0,tz);
 729
 730             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 731             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 732             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 733             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 734
 735             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 736
 737             /* Inner loop uses 178 flops */
 738         }
 739
 740         /* End of innermost loop */
 741
 742         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 743                                               f+i_coord_offset,fshift+i_shift_offset);
 744
 745         ggid                        = gid[iidx];
 746         /* Update potential energies */
 747         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 748         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 749
 750         /* Increment number of inner iterations */
 751         inneriter                  += j_index_end - j_index_start;
 752
 753         /* Outer loop uses 26 flops */
 754     }
 755
 756     /* Increment number of outer iterations */
 757     outeriter        += nri;
 758
 759     /* Update outer/inner flops */
 760
 761     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*178);
 762 }
 763 /*
 764  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse4_1_single
 765  * Electrostatics interaction: Ewald
 766  * VdW interaction:            LJEwald
 767  * Geometry:                   Water4-Particle
 768  * Calculate force/pot:        Force
 769  */
 770 void
 771 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse4_1_single
 772                     (t_nblist                    * gmx_restrict       nlist,
 773                      rvec                        * gmx_restrict          xx,
 774                      rvec                        * gmx_restrict          ff,
 775                      struct t_forcerec           * gmx_restrict          fr,
 776                      t_mdatoms                   * gmx_restrict     mdatoms,
 777                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 778                      t_nrnb                      * gmx_restrict        nrnb)
 779 {
 780     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 781      * just 0 for non-waters.
 782      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 783      * jnr indices corresponding to data put in the four positions in the SIMD register.
 784      */
 785     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 786     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 787     int              jnrA,jnrB,jnrC,jnrD;
 788     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 789     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 790     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 791     real             rcutoff_scalar;
 792     real             *shiftvec,*fshift,*x,*f;
 793     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 794     real             scratch[4*DIM];
 795     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 796     int              vdwioffset0;
 797     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 798     int              vdwioffset1;
 799     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 800     int              vdwioffset2;
 801     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 802     int              vdwioffset3;
 803     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 804     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 805     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 806     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 807     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 808     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 809     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 810     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 811     real             *charge;
 812     int              nvdwtype;
 813     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 814     int              *vdwtype;
 815     real             *vdwparam;
 816     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 817     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 818     __m128           c6grid_00;
 819     __m128           c6grid_10;
 820     __m128           c6grid_20;
 821     __m128           c6grid_30;
 822     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 823     real             *vdwgridparam;
 824     __m128           one_half  = _mm_set1_ps(0.5);
 825     __m128           minus_one = _mm_set1_ps(-1.0);
 826     __m128i          ewitab;
 827     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 828     real             *ewtab;
 829     __m128           dummy_mask,cutoff_mask;
 830     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 831     __m128           one     = _mm_set1_ps(1.0);
 832     __m128           two     = _mm_set1_ps(2.0);
 833     x                = xx[0];
 834     f                = ff[0];
 835
 836     nri              = nlist->nri;
 837     iinr             = nlist->iinr;
 838     jindex           = nlist->jindex;
 839     jjnr             = nlist->jjnr;
 840     shiftidx         = nlist->shift;
 841     gid              = nlist->gid;
 842     shiftvec         = fr->shift_vec[0];
 843     fshift           = fr->fshift[0];
 844     facel            = _mm_set1_ps(fr->ic->epsfac);
 845     charge           = mdatoms->chargeA;
 846     nvdwtype         = fr->ntype;
 847     vdwparam         = fr->nbfp;
 848     vdwtype          = mdatoms->typeA;
 849     vdwgridparam     = fr->ljpme_c6grid;
 850     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 851     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 852     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 853
 854     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 855     ewtab            = fr->ic->tabq_coul_F;
 856     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 857     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 858
 859     /* Setup water-specific parameters */
 860     inr              = nlist->iinr[0];
 861     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 862     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 863     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 864     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 865
 866     /* Avoid stupid compiler warnings */
 867     jnrA = jnrB = jnrC = jnrD = 0;
 868     j_coord_offsetA = 0;
 869     j_coord_offsetB = 0;
 870     j_coord_offsetC = 0;
 871     j_coord_offsetD = 0;
 872
 873     outeriter        = 0;
 874     inneriter        = 0;
 875
 876     for(iidx=0;iidx<4*DIM;iidx++)
 877     {
 878         scratch[iidx] = 0.0;
 879     }
 880
 881     /* Start outer loop over neighborlists */
 882     for(iidx=0; iidx<nri; iidx++)
 883     {
 884         /* Load shift vector for this list */
 885         i_shift_offset   = DIM*shiftidx[iidx];
 886
 887         /* Load limits for loop over neighbors */
 888         j_index_start    = jindex[iidx];
 889         j_index_end      = jindex[iidx+1];
 890
 891         /* Get outer coordinate index */
 892         inr              = iinr[iidx];
 893         i_coord_offset   = DIM*inr;
 894
 895         /* Load i particle coords and add shift vector */
 896         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 897                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 898
 899         fix0             = _mm_setzero_ps();
 900         fiy0             = _mm_setzero_ps();
 901         fiz0             = _mm_setzero_ps();
 902         fix1             = _mm_setzero_ps();
 903         fiy1             = _mm_setzero_ps();
 904         fiz1             = _mm_setzero_ps();
 905         fix2             = _mm_setzero_ps();
 906         fiy2             = _mm_setzero_ps();
 907         fiz2             = _mm_setzero_ps();
 908         fix3             = _mm_setzero_ps();
 909         fiy3             = _mm_setzero_ps();
 910         fiz3             = _mm_setzero_ps();
 911
 912         /* Start inner kernel loop */
 913         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 914         {
 915
 916             /* Get j neighbor index, and coordinate index */
 917             jnrA             = jjnr[jidx];
 918             jnrB             = jjnr[jidx+1];
 919             jnrC             = jjnr[jidx+2];
 920             jnrD             = jjnr[jidx+3];
 921             j_coord_offsetA  = DIM*jnrA;
 922             j_coord_offsetB  = DIM*jnrB;
 923             j_coord_offsetC  = DIM*jnrC;
 924             j_coord_offsetD  = DIM*jnrD;
 925
 926             /* load j atom coordinates */
 927             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 928                                               x+j_coord_offsetC,x+j_coord_offsetD,
 929                                               &jx0,&jy0,&jz0);
 930
 931             /* Calculate displacement vector */
 932             dx00             = _mm_sub_ps(ix0,jx0);
 933             dy00             = _mm_sub_ps(iy0,jy0);
 934             dz00             = _mm_sub_ps(iz0,jz0);
 935             dx10             = _mm_sub_ps(ix1,jx0);
 936             dy10             = _mm_sub_ps(iy1,jy0);
 937             dz10             = _mm_sub_ps(iz1,jz0);
 938             dx20             = _mm_sub_ps(ix2,jx0);
 939             dy20             = _mm_sub_ps(iy2,jy0);
 940             dz20             = _mm_sub_ps(iz2,jz0);
 941             dx30             = _mm_sub_ps(ix3,jx0);
 942             dy30             = _mm_sub_ps(iy3,jy0);
 943             dz30             = _mm_sub_ps(iz3,jz0);
 944
 945             /* Calculate squared distance and things based on it */
 946             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 947             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 948             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 949             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 950
 951             rinv00           = sse41_invsqrt_f(rsq00);
 952             rinv10           = sse41_invsqrt_f(rsq10);
 953             rinv20           = sse41_invsqrt_f(rsq20);
 954             rinv30           = sse41_invsqrt_f(rsq30);
 955
 956             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 957             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 958             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 959             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 960
 961             /* Load parameters for j particles */
 962             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 963                                                               charge+jnrC+0,charge+jnrD+0);
 964             vdwjidx0A        = 2*vdwtype[jnrA+0];
 965             vdwjidx0B        = 2*vdwtype[jnrB+0];
 966             vdwjidx0C        = 2*vdwtype[jnrC+0];
 967             vdwjidx0D        = 2*vdwtype[jnrD+0];
 968
 969             fjx0             = _mm_setzero_ps();
 970             fjy0             = _mm_setzero_ps();
 971             fjz0             = _mm_setzero_ps();
 972
 973             /**************************
 974              * CALCULATE INTERACTIONS *
 975              **************************/
 976
 977             r00              = _mm_mul_ps(rsq00,rinv00);
 978
 979             /* Compute parameters for interactions between i and j atoms */
 980             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 981                                          vdwparam+vdwioffset0+vdwjidx0B,
 982                                          vdwparam+vdwioffset0+vdwjidx0C,
 983                                          vdwparam+vdwioffset0+vdwjidx0D,
 984                                          &c6_00,&c12_00);
 985
 986             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 987                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 988                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 989                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 990
 991             /* Analytical LJ-PME */
 992             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 993             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 994             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 995             exponent         = sse41_exp_f(ewcljrsq);
 996             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 997             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 998             /* f6A = 6 * C6grid * (1 - poly) */
 999             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1000             /* f6B = C6grid * exponent * beta^6 */
1001             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1002             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1003             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);
1004
1005             fscal            = fvdw;
1006
1007             /* Calculate temporary vectorial force */
1008             tx               = _mm_mul_ps(fscal,dx00);
1009             ty               = _mm_mul_ps(fscal,dy00);
1010             tz               = _mm_mul_ps(fscal,dz00);
1011
1012             /* Update vectorial force */
1013             fix0             = _mm_add_ps(fix0,tx);
1014             fiy0             = _mm_add_ps(fiy0,ty);
1015             fiz0             = _mm_add_ps(fiz0,tz);
1016
1017             fjx0             = _mm_add_ps(fjx0,tx);
1018             fjy0             = _mm_add_ps(fjy0,ty);
1019             fjz0             = _mm_add_ps(fjz0,tz);
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             r10              = _mm_mul_ps(rsq10,rinv10);
1026
1027             /* Compute parameters for interactions between i and j atoms */
1028             qq10             = _mm_mul_ps(iq1,jq0);
1029
1030             /* EWALD ELECTROSTATICS */
1031
1032             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1033             ewrt             = _mm_mul_ps(r10,ewtabscale);
1034             ewitab           = _mm_cvttps_epi32(ewrt);
1035             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1036             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1037                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1038                                          &ewtabF,&ewtabFn);
1039             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1040             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1041
1042             fscal            = felec;
1043
1044             /* Calculate temporary vectorial force */
1045             tx               = _mm_mul_ps(fscal,dx10);
1046             ty               = _mm_mul_ps(fscal,dy10);
1047             tz               = _mm_mul_ps(fscal,dz10);
1048
1049             /* Update vectorial force */
1050             fix1             = _mm_add_ps(fix1,tx);
1051             fiy1             = _mm_add_ps(fiy1,ty);
1052             fiz1             = _mm_add_ps(fiz1,tz);
1053
1054             fjx0             = _mm_add_ps(fjx0,tx);
1055             fjy0             = _mm_add_ps(fjy0,ty);
1056             fjz0             = _mm_add_ps(fjz0,tz);
1057
1058             /**************************
1059              * CALCULATE INTERACTIONS *
1060              **************************/
1061
1062             r20              = _mm_mul_ps(rsq20,rinv20);
1063
1064             /* Compute parameters for interactions between i and j atoms */
1065             qq20             = _mm_mul_ps(iq2,jq0);
1066
1067             /* EWALD ELECTROSTATICS */
1068
1069             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1070             ewrt             = _mm_mul_ps(r20,ewtabscale);
1071             ewitab           = _mm_cvttps_epi32(ewrt);
1072             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1073             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1074                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1075                                          &ewtabF,&ewtabFn);
1076             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1077             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1078
1079             fscal            = felec;
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm_mul_ps(fscal,dx20);
1083             ty               = _mm_mul_ps(fscal,dy20);
1084             tz               = _mm_mul_ps(fscal,dz20);
1085
1086             /* Update vectorial force */
1087             fix2             = _mm_add_ps(fix2,tx);
1088             fiy2             = _mm_add_ps(fiy2,ty);
1089             fiz2             = _mm_add_ps(fiz2,tz);
1090
1091             fjx0             = _mm_add_ps(fjx0,tx);
1092             fjy0             = _mm_add_ps(fjy0,ty);
1093             fjz0             = _mm_add_ps(fjz0,tz);
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             r30              = _mm_mul_ps(rsq30,rinv30);
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq30             = _mm_mul_ps(iq3,jq0);
1103
1104             /* EWALD ELECTROSTATICS */
1105
1106             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1107             ewrt             = _mm_mul_ps(r30,ewtabscale);
1108             ewitab           = _mm_cvttps_epi32(ewrt);
1109             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1110             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1111                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1112                                          &ewtabF,&ewtabFn);
1113             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1114             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1115
1116             fscal            = felec;
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm_mul_ps(fscal,dx30);
1120             ty               = _mm_mul_ps(fscal,dy30);
1121             tz               = _mm_mul_ps(fscal,dz30);
1122
1123             /* Update vectorial force */
1124             fix3             = _mm_add_ps(fix3,tx);
1125             fiy3             = _mm_add_ps(fiy3,ty);
1126             fiz3             = _mm_add_ps(fiz3,tz);
1127
1128             fjx0             = _mm_add_ps(fjx0,tx);
1129             fjy0             = _mm_add_ps(fjy0,ty);
1130             fjz0             = _mm_add_ps(fjz0,tz);
1131
1132             fjptrA             = f+j_coord_offsetA;
1133             fjptrB             = f+j_coord_offsetB;
1134             fjptrC             = f+j_coord_offsetC;
1135             fjptrD             = f+j_coord_offsetD;
1136
1137             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1138
1139             /* Inner loop uses 154 flops */
1140         }
1141
1142         if(jidx<j_index_end)
1143         {
1144
1145             /* Get j neighbor index, and coordinate index */
1146             jnrlistA         = jjnr[jidx];
1147             jnrlistB         = jjnr[jidx+1];
1148             jnrlistC         = jjnr[jidx+2];
1149             jnrlistD         = jjnr[jidx+3];
1150             /* Sign of each element will be negative for non-real atoms.
1151              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1152              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1153              */
1154             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1155             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1156             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1157             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1158             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1159             j_coord_offsetA  = DIM*jnrA;
1160             j_coord_offsetB  = DIM*jnrB;
1161             j_coord_offsetC  = DIM*jnrC;
1162             j_coord_offsetD  = DIM*jnrD;
1163
1164             /* load j atom coordinates */
1165             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1166                                               x+j_coord_offsetC,x+j_coord_offsetD,
1167                                               &jx0,&jy0,&jz0);
1168
1169             /* Calculate displacement vector */
1170             dx00             = _mm_sub_ps(ix0,jx0);
1171             dy00             = _mm_sub_ps(iy0,jy0);
1172             dz00             = _mm_sub_ps(iz0,jz0);
1173             dx10             = _mm_sub_ps(ix1,jx0);
1174             dy10             = _mm_sub_ps(iy1,jy0);
1175             dz10             = _mm_sub_ps(iz1,jz0);
1176             dx20             = _mm_sub_ps(ix2,jx0);
1177             dy20             = _mm_sub_ps(iy2,jy0);
1178             dz20             = _mm_sub_ps(iz2,jz0);
1179             dx30             = _mm_sub_ps(ix3,jx0);
1180             dy30             = _mm_sub_ps(iy3,jy0);
1181             dz30             = _mm_sub_ps(iz3,jz0);
1182
1183             /* Calculate squared distance and things based on it */
1184             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1185             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1186             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1187             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1188
1189             rinv00           = sse41_invsqrt_f(rsq00);
1190             rinv10           = sse41_invsqrt_f(rsq10);
1191             rinv20           = sse41_invsqrt_f(rsq20);
1192             rinv30           = sse41_invsqrt_f(rsq30);
1193
1194             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1195             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1196             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1197             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1198
1199             /* Load parameters for j particles */
1200             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1201                                                               charge+jnrC+0,charge+jnrD+0);
1202             vdwjidx0A        = 2*vdwtype[jnrA+0];
1203             vdwjidx0B        = 2*vdwtype[jnrB+0];
1204             vdwjidx0C        = 2*vdwtype[jnrC+0];
1205             vdwjidx0D        = 2*vdwtype[jnrD+0];
1206
1207             fjx0             = _mm_setzero_ps();
1208             fjy0             = _mm_setzero_ps();
1209             fjz0             = _mm_setzero_ps();
1210
1211             /**************************
1212              * CALCULATE INTERACTIONS *
1213              **************************/
1214
1215             r00              = _mm_mul_ps(rsq00,rinv00);
1216             r00              = _mm_andnot_ps(dummy_mask,r00);
1217
1218             /* Compute parameters for interactions between i and j atoms */
1219             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1220                                          vdwparam+vdwioffset0+vdwjidx0B,
1221                                          vdwparam+vdwioffset0+vdwjidx0C,
1222                                          vdwparam+vdwioffset0+vdwjidx0D,
1223                                          &c6_00,&c12_00);
1224
1225             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1226                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1227                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1228                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1229
1230             /* Analytical LJ-PME */
1231             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1232             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1233             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1234             exponent         = sse41_exp_f(ewcljrsq);
1235             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1236             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1237             /* f6A = 6 * C6grid * (1 - poly) */
1238             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1239             /* f6B = C6grid * exponent * beta^6 */
1240             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1241             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1242             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);
1243
1244             fscal            = fvdw;
1245
1246             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1247
1248             /* Calculate temporary vectorial force */
1249             tx               = _mm_mul_ps(fscal,dx00);
1250             ty               = _mm_mul_ps(fscal,dy00);
1251             tz               = _mm_mul_ps(fscal,dz00);
1252
1253             /* Update vectorial force */
1254             fix0             = _mm_add_ps(fix0,tx);
1255             fiy0             = _mm_add_ps(fiy0,ty);
1256             fiz0             = _mm_add_ps(fiz0,tz);
1257
1258             fjx0             = _mm_add_ps(fjx0,tx);
1259             fjy0             = _mm_add_ps(fjy0,ty);
1260             fjz0             = _mm_add_ps(fjz0,tz);
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             r10              = _mm_mul_ps(rsq10,rinv10);
1267             r10              = _mm_andnot_ps(dummy_mask,r10);
1268
1269             /* Compute parameters for interactions between i and j atoms */
1270             qq10             = _mm_mul_ps(iq1,jq0);
1271
1272             /* EWALD ELECTROSTATICS */
1273
1274             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1275             ewrt             = _mm_mul_ps(r10,ewtabscale);
1276             ewitab           = _mm_cvttps_epi32(ewrt);
1277             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1278             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1279                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1280                                          &ewtabF,&ewtabFn);
1281             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1282             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1283
1284             fscal            = felec;
1285
1286             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1287
1288             /* Calculate temporary vectorial force */
1289             tx               = _mm_mul_ps(fscal,dx10);
1290             ty               = _mm_mul_ps(fscal,dy10);
1291             tz               = _mm_mul_ps(fscal,dz10);
1292
1293             /* Update vectorial force */
1294             fix1             = _mm_add_ps(fix1,tx);
1295             fiy1             = _mm_add_ps(fiy1,ty);
1296             fiz1             = _mm_add_ps(fiz1,tz);
1297
1298             fjx0             = _mm_add_ps(fjx0,tx);
1299             fjy0             = _mm_add_ps(fjy0,ty);
1300             fjz0             = _mm_add_ps(fjz0,tz);
1301
1302             /**************************
1303              * CALCULATE INTERACTIONS *
1304              **************************/
1305
1306             r20              = _mm_mul_ps(rsq20,rinv20);
1307             r20              = _mm_andnot_ps(dummy_mask,r20);
1308
1309             /* Compute parameters for interactions between i and j atoms */
1310             qq20             = _mm_mul_ps(iq2,jq0);
1311
1312             /* EWALD ELECTROSTATICS */
1313
1314             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1315             ewrt             = _mm_mul_ps(r20,ewtabscale);
1316             ewitab           = _mm_cvttps_epi32(ewrt);
1317             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1318             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1319                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1320                                          &ewtabF,&ewtabFn);
1321             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1322             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1323
1324             fscal            = felec;
1325
1326             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1327
1328             /* Calculate temporary vectorial force */
1329             tx               = _mm_mul_ps(fscal,dx20);
1330             ty               = _mm_mul_ps(fscal,dy20);
1331             tz               = _mm_mul_ps(fscal,dz20);
1332
1333             /* Update vectorial force */
1334             fix2             = _mm_add_ps(fix2,tx);
1335             fiy2             = _mm_add_ps(fiy2,ty);
1336             fiz2             = _mm_add_ps(fiz2,tz);
1337
1338             fjx0             = _mm_add_ps(fjx0,tx);
1339             fjy0             = _mm_add_ps(fjy0,ty);
1340             fjz0             = _mm_add_ps(fjz0,tz);
1341
1342             /**************************
1343              * CALCULATE INTERACTIONS *
1344              **************************/
1345
1346             r30              = _mm_mul_ps(rsq30,rinv30);
1347             r30              = _mm_andnot_ps(dummy_mask,r30);
1348
1349             /* Compute parameters for interactions between i and j atoms */
1350             qq30             = _mm_mul_ps(iq3,jq0);
1351
1352             /* EWALD ELECTROSTATICS */
1353
1354             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1355             ewrt             = _mm_mul_ps(r30,ewtabscale);
1356             ewitab           = _mm_cvttps_epi32(ewrt);
1357             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1358             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1359                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1360                                          &ewtabF,&ewtabFn);
1361             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1362             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1363
1364             fscal            = felec;
1365
1366             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1367
1368             /* Calculate temporary vectorial force */
1369             tx               = _mm_mul_ps(fscal,dx30);
1370             ty               = _mm_mul_ps(fscal,dy30);
1371             tz               = _mm_mul_ps(fscal,dz30);
1372
1373             /* Update vectorial force */
1374             fix3             = _mm_add_ps(fix3,tx);
1375             fiy3             = _mm_add_ps(fiy3,ty);
1376             fiz3             = _mm_add_ps(fiz3,tz);
1377
1378             fjx0             = _mm_add_ps(fjx0,tx);
1379             fjy0             = _mm_add_ps(fjy0,ty);
1380             fjz0             = _mm_add_ps(fjz0,tz);
1381
1382             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1383             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1384             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1385             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1386
1387             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1388
1389             /* Inner loop uses 158 flops */
1390         }
1391
1392         /* End of innermost loop */
1393
1394         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1395                                               f+i_coord_offset,fshift+i_shift_offset);
1396
1397         /* Increment number of inner iterations */
1398         inneriter                  += j_index_end - j_index_start;
1399
1400         /* Outer loop uses 24 flops */
1401     }
1402
1403     /* Increment number of outer iterations */
1404     outeriter        += nri;
1405
1406     /* Update outer/inner flops */
1407
1408     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*158);
1409 }