src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_sse4_1_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_GeomW3P1_VF_sse4_1_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128           c6grid_00;
 102     __m128           c6grid_10;
 103     __m128           c6grid_20;
 104     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 105     real             *vdwgridparam;
 106     __m128           one_half  = _mm_set1_ps(0.5);
 107     __m128           minus_one = _mm_set1_ps(-1.0);
 108     __m128i          ewitab;
 109     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     real             *ewtab;
 111     __m128           dummy_mask,cutoff_mask;
 112     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 113     __m128           one     = _mm_set1_ps(1.0);
 114     __m128           two     = _mm_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_ps(fr->ic->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131     vdwgridparam     = fr->ljpme_c6grid;
 132     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 133     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 134     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 135
 136     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 144     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 145     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 146     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = jnrC = jnrD = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152     j_coord_offsetC = 0;
 153     j_coord_offsetD = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 180
 181         fix0             = _mm_setzero_ps();
 182         fiy0             = _mm_setzero_ps();
 183         fiz0             = _mm_setzero_ps();
 184         fix1             = _mm_setzero_ps();
 185         fiy1             = _mm_setzero_ps();
 186         fiz1             = _mm_setzero_ps();
 187         fix2             = _mm_setzero_ps();
 188         fiy2             = _mm_setzero_ps();
 189         fiz2             = _mm_setzero_ps();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_ps();
 193         vvdwsum          = _mm_setzero_ps();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208
 209             /* load j atom coordinates */
 210             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                               x+j_coord_offsetC,x+j_coord_offsetD,
 212                                               &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm_sub_ps(ix0,jx0);
 216             dy00             = _mm_sub_ps(iy0,jy0);
 217             dz00             = _mm_sub_ps(iz0,jz0);
 218             dx10             = _mm_sub_ps(ix1,jx0);
 219             dy10             = _mm_sub_ps(iy1,jy0);
 220             dz10             = _mm_sub_ps(iz1,jz0);
 221             dx20             = _mm_sub_ps(ix2,jx0);
 222             dy20             = _mm_sub_ps(iy2,jy0);
 223             dz20             = _mm_sub_ps(iz2,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 227             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 228             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 229
 230             rinv00           = sse41_invsqrt_f(rsq00);
 231             rinv10           = sse41_invsqrt_f(rsq10);
 232             rinv20           = sse41_invsqrt_f(rsq20);
 233
 234             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 235             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 236             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 237
 238             /* Load parameters for j particles */
 239             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 240                                                               charge+jnrC+0,charge+jnrD+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243             vdwjidx0C        = 2*vdwtype[jnrC+0];
 244             vdwjidx0D        = 2*vdwtype[jnrD+0];
 245
 246             fjx0             = _mm_setzero_ps();
 247             fjy0             = _mm_setzero_ps();
 248             fjz0             = _mm_setzero_ps();
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             r00              = _mm_mul_ps(rsq00,rinv00);
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq00             = _mm_mul_ps(iq0,jq0);
 258             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 259                                          vdwparam+vdwioffset0+vdwjidx0B,
 260                                          vdwparam+vdwioffset0+vdwjidx0C,
 261                                          vdwparam+vdwioffset0+vdwjidx0D,
 262                                          &c6_00,&c12_00);
 263
 264             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 265                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 266                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 267                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 268
 269             /* EWALD ELECTROSTATICS */
 270
 271             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 272             ewrt             = _mm_mul_ps(r00,ewtabscale);
 273             ewitab           = _mm_cvttps_epi32(ewrt);
 274             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 275             ewitab           = _mm_slli_epi32(ewitab,2);
 276             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 277             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 278             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 279             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 280             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 281             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 282             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 283             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 284             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 285
 286             /* Analytical LJ-PME */
 287             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 288             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 289             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 290             exponent         = sse41_exp_f(ewcljrsq);
 291             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 292             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 293             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 294             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 295             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 296             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 297             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 298             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);
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velecsum         = _mm_add_ps(velecsum,velec);
 302             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 303
 304             fscal            = _mm_add_ps(felec,fvdw);
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm_mul_ps(fscal,dx00);
 308             ty               = _mm_mul_ps(fscal,dy00);
 309             tz               = _mm_mul_ps(fscal,dz00);
 310
 311             /* Update vectorial force */
 312             fix0             = _mm_add_ps(fix0,tx);
 313             fiy0             = _mm_add_ps(fiy0,ty);
 314             fiz0             = _mm_add_ps(fiz0,tz);
 315
 316             fjx0             = _mm_add_ps(fjx0,tx);
 317             fjy0             = _mm_add_ps(fjy0,ty);
 318             fjz0             = _mm_add_ps(fjz0,tz);
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             r10              = _mm_mul_ps(rsq10,rinv10);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq10             = _mm_mul_ps(iq1,jq0);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 332             ewrt             = _mm_mul_ps(r10,ewtabscale);
 333             ewitab           = _mm_cvttps_epi32(ewrt);
 334             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 335             ewitab           = _mm_slli_epi32(ewitab,2);
 336             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 337             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 338             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 339             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 340             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 341             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 342             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 343             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 344             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm_mul_ps(fscal,dx10);
 353             ty               = _mm_mul_ps(fscal,dy10);
 354             tz               = _mm_mul_ps(fscal,dz10);
 355
 356             /* Update vectorial force */
 357             fix1             = _mm_add_ps(fix1,tx);
 358             fiy1             = _mm_add_ps(fiy1,ty);
 359             fiz1             = _mm_add_ps(fiz1,tz);
 360
 361             fjx0             = _mm_add_ps(fjx0,tx);
 362             fjy0             = _mm_add_ps(fjy0,ty);
 363             fjz0             = _mm_add_ps(fjz0,tz);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             r20              = _mm_mul_ps(rsq20,rinv20);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq20             = _mm_mul_ps(iq2,jq0);
 373
 374             /* EWALD ELECTROSTATICS */
 375
 376             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 377             ewrt             = _mm_mul_ps(r20,ewtabscale);
 378             ewitab           = _mm_cvttps_epi32(ewrt);
 379             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 380             ewitab           = _mm_slli_epi32(ewitab,2);
 381             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 382             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 383             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 384             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 385             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 386             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 387             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 388             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 389             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velecsum         = _mm_add_ps(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm_mul_ps(fscal,dx20);
 398             ty               = _mm_mul_ps(fscal,dy20);
 399             tz               = _mm_mul_ps(fscal,dz20);
 400
 401             /* Update vectorial force */
 402             fix2             = _mm_add_ps(fix2,tx);
 403             fiy2             = _mm_add_ps(fiy2,ty);
 404             fiz2             = _mm_add_ps(fiz2,tz);
 405
 406             fjx0             = _mm_add_ps(fjx0,tx);
 407             fjy0             = _mm_add_ps(fjy0,ty);
 408             fjz0             = _mm_add_ps(fjz0,tz);
 409
 410             fjptrA             = f+j_coord_offsetA;
 411             fjptrB             = f+j_coord_offsetB;
 412             fjptrC             = f+j_coord_offsetC;
 413             fjptrD             = f+j_coord_offsetD;
 414
 415             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 416
 417             /* Inner loop uses 151 flops */
 418         }
 419
 420         if(jidx<j_index_end)
 421         {
 422
 423             /* Get j neighbor index, and coordinate index */
 424             jnrlistA         = jjnr[jidx];
 425             jnrlistB         = jjnr[jidx+1];
 426             jnrlistC         = jjnr[jidx+2];
 427             jnrlistD         = jjnr[jidx+3];
 428             /* Sign of each element will be negative for non-real atoms.
 429              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 430              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 431              */
 432             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 433             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 434             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 435             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 436             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 437             j_coord_offsetA  = DIM*jnrA;
 438             j_coord_offsetB  = DIM*jnrB;
 439             j_coord_offsetC  = DIM*jnrC;
 440             j_coord_offsetD  = DIM*jnrD;
 441
 442             /* load j atom coordinates */
 443             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 444                                               x+j_coord_offsetC,x+j_coord_offsetD,
 445                                               &jx0,&jy0,&jz0);
 446
 447             /* Calculate displacement vector */
 448             dx00             = _mm_sub_ps(ix0,jx0);
 449             dy00             = _mm_sub_ps(iy0,jy0);
 450             dz00             = _mm_sub_ps(iz0,jz0);
 451             dx10             = _mm_sub_ps(ix1,jx0);
 452             dy10             = _mm_sub_ps(iy1,jy0);
 453             dz10             = _mm_sub_ps(iz1,jz0);
 454             dx20             = _mm_sub_ps(ix2,jx0);
 455             dy20             = _mm_sub_ps(iy2,jy0);
 456             dz20             = _mm_sub_ps(iz2,jz0);
 457
 458             /* Calculate squared distance and things based on it */
 459             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 460             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 461             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 462
 463             rinv00           = sse41_invsqrt_f(rsq00);
 464             rinv10           = sse41_invsqrt_f(rsq10);
 465             rinv20           = sse41_invsqrt_f(rsq20);
 466
 467             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 468             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 469             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 470
 471             /* Load parameters for j particles */
 472             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 473                                                               charge+jnrC+0,charge+jnrD+0);
 474             vdwjidx0A        = 2*vdwtype[jnrA+0];
 475             vdwjidx0B        = 2*vdwtype[jnrB+0];
 476             vdwjidx0C        = 2*vdwtype[jnrC+0];
 477             vdwjidx0D        = 2*vdwtype[jnrD+0];
 478
 479             fjx0             = _mm_setzero_ps();
 480             fjy0             = _mm_setzero_ps();
 481             fjz0             = _mm_setzero_ps();
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             r00              = _mm_mul_ps(rsq00,rinv00);
 488             r00              = _mm_andnot_ps(dummy_mask,r00);
 489
 490             /* Compute parameters for interactions between i and j atoms */
 491             qq00             = _mm_mul_ps(iq0,jq0);
 492             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 493                                          vdwparam+vdwioffset0+vdwjidx0B,
 494                                          vdwparam+vdwioffset0+vdwjidx0C,
 495                                          vdwparam+vdwioffset0+vdwjidx0D,
 496                                          &c6_00,&c12_00);
 497
 498             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 499                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 500                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 501                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 502
 503             /* EWALD ELECTROSTATICS */
 504
 505             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 506             ewrt             = _mm_mul_ps(r00,ewtabscale);
 507             ewitab           = _mm_cvttps_epi32(ewrt);
 508             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 509             ewitab           = _mm_slli_epi32(ewitab,2);
 510             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 511             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 512             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 513             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 514             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 515             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 516             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 517             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 518             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 519
 520             /* Analytical LJ-PME */
 521             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 522             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 523             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 524             exponent         = sse41_exp_f(ewcljrsq);
 525             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 526             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 527             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 528             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 529             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 530             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 531             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 532             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);
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm_andnot_ps(dummy_mask,velec);
 536             velecsum         = _mm_add_ps(velecsum,velec);
 537             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 538             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 539
 540             fscal            = _mm_add_ps(felec,fvdw);
 541
 542             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 543
 544             /* Calculate temporary vectorial force */
 545             tx               = _mm_mul_ps(fscal,dx00);
 546             ty               = _mm_mul_ps(fscal,dy00);
 547             tz               = _mm_mul_ps(fscal,dz00);
 548
 549             /* Update vectorial force */
 550             fix0             = _mm_add_ps(fix0,tx);
 551             fiy0             = _mm_add_ps(fiy0,ty);
 552             fiz0             = _mm_add_ps(fiz0,tz);
 553
 554             fjx0             = _mm_add_ps(fjx0,tx);
 555             fjy0             = _mm_add_ps(fjy0,ty);
 556             fjz0             = _mm_add_ps(fjz0,tz);
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             r10              = _mm_mul_ps(rsq10,rinv10);
 563             r10              = _mm_andnot_ps(dummy_mask,r10);
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             qq10             = _mm_mul_ps(iq1,jq0);
 567
 568             /* EWALD ELECTROSTATICS */
 569
 570             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 571             ewrt             = _mm_mul_ps(r10,ewtabscale);
 572             ewitab           = _mm_cvttps_epi32(ewrt);
 573             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 574             ewitab           = _mm_slli_epi32(ewitab,2);
 575             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 576             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 577             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 578             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 579             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 580             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 581             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 582             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 583             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 584
 585             /* Update potential sum for this i atom from the interaction with this j atom. */
 586             velec            = _mm_andnot_ps(dummy_mask,velec);
 587             velecsum         = _mm_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm_mul_ps(fscal,dx10);
 595             ty               = _mm_mul_ps(fscal,dy10);
 596             tz               = _mm_mul_ps(fscal,dz10);
 597
 598             /* Update vectorial force */
 599             fix1             = _mm_add_ps(fix1,tx);
 600             fiy1             = _mm_add_ps(fiy1,ty);
 601             fiz1             = _mm_add_ps(fiz1,tz);
 602
 603             fjx0             = _mm_add_ps(fjx0,tx);
 604             fjy0             = _mm_add_ps(fjy0,ty);
 605             fjz0             = _mm_add_ps(fjz0,tz);
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             r20              = _mm_mul_ps(rsq20,rinv20);
 612             r20              = _mm_andnot_ps(dummy_mask,r20);
 613
 614             /* Compute parameters for interactions between i and j atoms */
 615             qq20             = _mm_mul_ps(iq2,jq0);
 616
 617             /* EWALD ELECTROSTATICS */
 618
 619             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 620             ewrt             = _mm_mul_ps(r20,ewtabscale);
 621             ewitab           = _mm_cvttps_epi32(ewrt);
 622             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 623             ewitab           = _mm_slli_epi32(ewitab,2);
 624             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 625             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 626             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 627             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 628             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 629             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 630             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 631             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 632             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             velec            = _mm_andnot_ps(dummy_mask,velec);
 636             velecsum         = _mm_add_ps(velecsum,velec);
 637
 638             fscal            = felec;
 639
 640             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 641
 642             /* Calculate temporary vectorial force */
 643             tx               = _mm_mul_ps(fscal,dx20);
 644             ty               = _mm_mul_ps(fscal,dy20);
 645             tz               = _mm_mul_ps(fscal,dz20);
 646
 647             /* Update vectorial force */
 648             fix2             = _mm_add_ps(fix2,tx);
 649             fiy2             = _mm_add_ps(fiy2,ty);
 650             fiz2             = _mm_add_ps(fiz2,tz);
 651
 652             fjx0             = _mm_add_ps(fjx0,tx);
 653             fjy0             = _mm_add_ps(fjy0,ty);
 654             fjz0             = _mm_add_ps(fjz0,tz);
 655
 656             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 657             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 658             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 659             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 660
 661             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 662
 663             /* Inner loop uses 154 flops */
 664         }
 665
 666         /* End of innermost loop */
 667
 668         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 669                                               f+i_coord_offset,fshift+i_shift_offset);
 670
 671         ggid                        = gid[iidx];
 672         /* Update potential energies */
 673         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 674         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 675
 676         /* Increment number of inner iterations */
 677         inneriter                  += j_index_end - j_index_start;
 678
 679         /* Outer loop uses 20 flops */
 680     }
 681
 682     /* Increment number of outer iterations */
 683     outeriter        += nri;
 684
 685     /* Update outer/inner flops */
 686
 687     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 688 }
 689 /*
 690  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse4_1_single
 691  * Electrostatics interaction: Ewald
 692  * VdW interaction:            LJEwald
 693  * Geometry:                   Water3-Particle
 694  * Calculate force/pot:        Force
 695  */
 696 void
 697 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse4_1_single
 698                     (t_nblist                    * gmx_restrict       nlist,
 699                      rvec                        * gmx_restrict          xx,
 700                      rvec                        * gmx_restrict          ff,
 701                      struct t_forcerec           * gmx_restrict          fr,
 702                      t_mdatoms                   * gmx_restrict     mdatoms,
 703                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 704                      t_nrnb                      * gmx_restrict        nrnb)
 705 {
 706     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 707      * just 0 for non-waters.
 708      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 709      * jnr indices corresponding to data put in the four positions in the SIMD register.
 710      */
 711     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 712     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 713     int              jnrA,jnrB,jnrC,jnrD;
 714     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 715     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 716     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 717     real             rcutoff_scalar;
 718     real             *shiftvec,*fshift,*x,*f;
 719     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 720     real             scratch[4*DIM];
 721     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 722     int              vdwioffset0;
 723     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 724     int              vdwioffset1;
 725     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 726     int              vdwioffset2;
 727     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 728     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 729     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 730     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 731     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 732     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 733     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 734     real             *charge;
 735     int              nvdwtype;
 736     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 737     int              *vdwtype;
 738     real             *vdwparam;
 739     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 740     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 741     __m128           c6grid_00;
 742     __m128           c6grid_10;
 743     __m128           c6grid_20;
 744     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 745     real             *vdwgridparam;
 746     __m128           one_half  = _mm_set1_ps(0.5);
 747     __m128           minus_one = _mm_set1_ps(-1.0);
 748     __m128i          ewitab;
 749     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 750     real             *ewtab;
 751     __m128           dummy_mask,cutoff_mask;
 752     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 753     __m128           one     = _mm_set1_ps(1.0);
 754     __m128           two     = _mm_set1_ps(2.0);
 755     x                = xx[0];
 756     f                = ff[0];
 757
 758     nri              = nlist->nri;
 759     iinr             = nlist->iinr;
 760     jindex           = nlist->jindex;
 761     jjnr             = nlist->jjnr;
 762     shiftidx         = nlist->shift;
 763     gid              = nlist->gid;
 764     shiftvec         = fr->shift_vec[0];
 765     fshift           = fr->fshift[0];
 766     facel            = _mm_set1_ps(fr->ic->epsfac);
 767     charge           = mdatoms->chargeA;
 768     nvdwtype         = fr->ntype;
 769     vdwparam         = fr->nbfp;
 770     vdwtype          = mdatoms->typeA;
 771     vdwgridparam     = fr->ljpme_c6grid;
 772     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 773     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 774     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 775
 776     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 777     ewtab            = fr->ic->tabq_coul_F;
 778     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 779     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 780
 781     /* Setup water-specific parameters */
 782     inr              = nlist->iinr[0];
 783     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 784     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 785     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 786     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 787
 788     /* Avoid stupid compiler warnings */
 789     jnrA = jnrB = jnrC = jnrD = 0;
 790     j_coord_offsetA = 0;
 791     j_coord_offsetB = 0;
 792     j_coord_offsetC = 0;
 793     j_coord_offsetD = 0;
 794
 795     outeriter        = 0;
 796     inneriter        = 0;
 797
 798     for(iidx=0;iidx<4*DIM;iidx++)
 799     {
 800         scratch[iidx] = 0.0;
 801     }
 802
 803     /* Start outer loop over neighborlists */
 804     for(iidx=0; iidx<nri; iidx++)
 805     {
 806         /* Load shift vector for this list */
 807         i_shift_offset   = DIM*shiftidx[iidx];
 808
 809         /* Load limits for loop over neighbors */
 810         j_index_start    = jindex[iidx];
 811         j_index_end      = jindex[iidx+1];
 812
 813         /* Get outer coordinate index */
 814         inr              = iinr[iidx];
 815         i_coord_offset   = DIM*inr;
 816
 817         /* Load i particle coords and add shift vector */
 818         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 819                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 820
 821         fix0             = _mm_setzero_ps();
 822         fiy0             = _mm_setzero_ps();
 823         fiz0             = _mm_setzero_ps();
 824         fix1             = _mm_setzero_ps();
 825         fiy1             = _mm_setzero_ps();
 826         fiz1             = _mm_setzero_ps();
 827         fix2             = _mm_setzero_ps();
 828         fiy2             = _mm_setzero_ps();
 829         fiz2             = _mm_setzero_ps();
 830
 831         /* Start inner kernel loop */
 832         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 833         {
 834
 835             /* Get j neighbor index, and coordinate index */
 836             jnrA             = jjnr[jidx];
 837             jnrB             = jjnr[jidx+1];
 838             jnrC             = jjnr[jidx+2];
 839             jnrD             = jjnr[jidx+3];
 840             j_coord_offsetA  = DIM*jnrA;
 841             j_coord_offsetB  = DIM*jnrB;
 842             j_coord_offsetC  = DIM*jnrC;
 843             j_coord_offsetD  = DIM*jnrD;
 844
 845             /* load j atom coordinates */
 846             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 847                                               x+j_coord_offsetC,x+j_coord_offsetD,
 848                                               &jx0,&jy0,&jz0);
 849
 850             /* Calculate displacement vector */
 851             dx00             = _mm_sub_ps(ix0,jx0);
 852             dy00             = _mm_sub_ps(iy0,jy0);
 853             dz00             = _mm_sub_ps(iz0,jz0);
 854             dx10             = _mm_sub_ps(ix1,jx0);
 855             dy10             = _mm_sub_ps(iy1,jy0);
 856             dz10             = _mm_sub_ps(iz1,jz0);
 857             dx20             = _mm_sub_ps(ix2,jx0);
 858             dy20             = _mm_sub_ps(iy2,jy0);
 859             dz20             = _mm_sub_ps(iz2,jz0);
 860
 861             /* Calculate squared distance and things based on it */
 862             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 863             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 864             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 865
 866             rinv00           = sse41_invsqrt_f(rsq00);
 867             rinv10           = sse41_invsqrt_f(rsq10);
 868             rinv20           = sse41_invsqrt_f(rsq20);
 869
 870             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 871             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 872             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 873
 874             /* Load parameters for j particles */
 875             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 876                                                               charge+jnrC+0,charge+jnrD+0);
 877             vdwjidx0A        = 2*vdwtype[jnrA+0];
 878             vdwjidx0B        = 2*vdwtype[jnrB+0];
 879             vdwjidx0C        = 2*vdwtype[jnrC+0];
 880             vdwjidx0D        = 2*vdwtype[jnrD+0];
 881
 882             fjx0             = _mm_setzero_ps();
 883             fjy0             = _mm_setzero_ps();
 884             fjz0             = _mm_setzero_ps();
 885
 886             /**************************
 887              * CALCULATE INTERACTIONS *
 888              **************************/
 889
 890             r00              = _mm_mul_ps(rsq00,rinv00);
 891
 892             /* Compute parameters for interactions between i and j atoms */
 893             qq00             = _mm_mul_ps(iq0,jq0);
 894             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 895                                          vdwparam+vdwioffset0+vdwjidx0B,
 896                                          vdwparam+vdwioffset0+vdwjidx0C,
 897                                          vdwparam+vdwioffset0+vdwjidx0D,
 898                                          &c6_00,&c12_00);
 899
 900             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 901                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 902                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 903                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 904
 905             /* EWALD ELECTROSTATICS */
 906
 907             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 908             ewrt             = _mm_mul_ps(r00,ewtabscale);
 909             ewitab           = _mm_cvttps_epi32(ewrt);
 910             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 911             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 912                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 913                                          &ewtabF,&ewtabFn);
 914             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 915             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 916
 917             /* Analytical LJ-PME */
 918             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 919             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 920             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 921             exponent         = sse41_exp_f(ewcljrsq);
 922             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 923             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 924             /* f6A = 6 * C6grid * (1 - poly) */
 925             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 926             /* f6B = C6grid * exponent * beta^6 */
 927             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 928             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 929             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);
 930
 931             fscal            = _mm_add_ps(felec,fvdw);
 932
 933             /* Calculate temporary vectorial force */
 934             tx               = _mm_mul_ps(fscal,dx00);
 935             ty               = _mm_mul_ps(fscal,dy00);
 936             tz               = _mm_mul_ps(fscal,dz00);
 937
 938             /* Update vectorial force */
 939             fix0             = _mm_add_ps(fix0,tx);
 940             fiy0             = _mm_add_ps(fiy0,ty);
 941             fiz0             = _mm_add_ps(fiz0,tz);
 942
 943             fjx0             = _mm_add_ps(fjx0,tx);
 944             fjy0             = _mm_add_ps(fjy0,ty);
 945             fjz0             = _mm_add_ps(fjz0,tz);
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             r10              = _mm_mul_ps(rsq10,rinv10);
 952
 953             /* Compute parameters for interactions between i and j atoms */
 954             qq10             = _mm_mul_ps(iq1,jq0);
 955
 956             /* EWALD ELECTROSTATICS */
 957
 958             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 959             ewrt             = _mm_mul_ps(r10,ewtabscale);
 960             ewitab           = _mm_cvttps_epi32(ewrt);
 961             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 962             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 963                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 964                                          &ewtabF,&ewtabFn);
 965             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 966             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 967
 968             fscal            = felec;
 969
 970             /* Calculate temporary vectorial force */
 971             tx               = _mm_mul_ps(fscal,dx10);
 972             ty               = _mm_mul_ps(fscal,dy10);
 973             tz               = _mm_mul_ps(fscal,dz10);
 974
 975             /* Update vectorial force */
 976             fix1             = _mm_add_ps(fix1,tx);
 977             fiy1             = _mm_add_ps(fiy1,ty);
 978             fiz1             = _mm_add_ps(fiz1,tz);
 979
 980             fjx0             = _mm_add_ps(fjx0,tx);
 981             fjy0             = _mm_add_ps(fjy0,ty);
 982             fjz0             = _mm_add_ps(fjz0,tz);
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             r20              = _mm_mul_ps(rsq20,rinv20);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq20             = _mm_mul_ps(iq2,jq0);
 992
 993             /* EWALD ELECTROSTATICS */
 994
 995             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 996             ewrt             = _mm_mul_ps(r20,ewtabscale);
 997             ewitab           = _mm_cvttps_epi32(ewrt);
 998             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 999             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1000                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1001                                          &ewtabF,&ewtabFn);
1002             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1003             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1004
1005             fscal            = felec;
1006
1007             /* Calculate temporary vectorial force */
1008             tx               = _mm_mul_ps(fscal,dx20);
1009             ty               = _mm_mul_ps(fscal,dy20);
1010             tz               = _mm_mul_ps(fscal,dz20);
1011
1012             /* Update vectorial force */
1013             fix2             = _mm_add_ps(fix2,tx);
1014             fiy2             = _mm_add_ps(fiy2,ty);
1015             fiz2             = _mm_add_ps(fiz2,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             fjptrA             = f+j_coord_offsetA;
1022             fjptrB             = f+j_coord_offsetB;
1023             fjptrC             = f+j_coord_offsetC;
1024             fjptrD             = f+j_coord_offsetD;
1025
1026             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1027
1028             /* Inner loop uses 131 flops */
1029         }
1030
1031         if(jidx<j_index_end)
1032         {
1033
1034             /* Get j neighbor index, and coordinate index */
1035             jnrlistA         = jjnr[jidx];
1036             jnrlistB         = jjnr[jidx+1];
1037             jnrlistC         = jjnr[jidx+2];
1038             jnrlistD         = jjnr[jidx+3];
1039             /* Sign of each element will be negative for non-real atoms.
1040              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1041              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1042              */
1043             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1044             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1045             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1046             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1047             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1048             j_coord_offsetA  = DIM*jnrA;
1049             j_coord_offsetB  = DIM*jnrB;
1050             j_coord_offsetC  = DIM*jnrC;
1051             j_coord_offsetD  = DIM*jnrD;
1052
1053             /* load j atom coordinates */
1054             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1055                                               x+j_coord_offsetC,x+j_coord_offsetD,
1056                                               &jx0,&jy0,&jz0);
1057
1058             /* Calculate displacement vector */
1059             dx00             = _mm_sub_ps(ix0,jx0);
1060             dy00             = _mm_sub_ps(iy0,jy0);
1061             dz00             = _mm_sub_ps(iz0,jz0);
1062             dx10             = _mm_sub_ps(ix1,jx0);
1063             dy10             = _mm_sub_ps(iy1,jy0);
1064             dz10             = _mm_sub_ps(iz1,jz0);
1065             dx20             = _mm_sub_ps(ix2,jx0);
1066             dy20             = _mm_sub_ps(iy2,jy0);
1067             dz20             = _mm_sub_ps(iz2,jz0);
1068
1069             /* Calculate squared distance and things based on it */
1070             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1071             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1072             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1073
1074             rinv00           = sse41_invsqrt_f(rsq00);
1075             rinv10           = sse41_invsqrt_f(rsq10);
1076             rinv20           = sse41_invsqrt_f(rsq20);
1077
1078             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1079             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1080             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1081
1082             /* Load parameters for j particles */
1083             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1084                                                               charge+jnrC+0,charge+jnrD+0);
1085             vdwjidx0A        = 2*vdwtype[jnrA+0];
1086             vdwjidx0B        = 2*vdwtype[jnrB+0];
1087             vdwjidx0C        = 2*vdwtype[jnrC+0];
1088             vdwjidx0D        = 2*vdwtype[jnrD+0];
1089
1090             fjx0             = _mm_setzero_ps();
1091             fjy0             = _mm_setzero_ps();
1092             fjz0             = _mm_setzero_ps();
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             r00              = _mm_mul_ps(rsq00,rinv00);
1099             r00              = _mm_andnot_ps(dummy_mask,r00);
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq00             = _mm_mul_ps(iq0,jq0);
1103             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1104                                          vdwparam+vdwioffset0+vdwjidx0B,
1105                                          vdwparam+vdwioffset0+vdwjidx0C,
1106                                          vdwparam+vdwioffset0+vdwjidx0D,
1107                                          &c6_00,&c12_00);
1108
1109             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1110                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1111                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1112                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1113
1114             /* EWALD ELECTROSTATICS */
1115
1116             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1117             ewrt             = _mm_mul_ps(r00,ewtabscale);
1118             ewitab           = _mm_cvttps_epi32(ewrt);
1119             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1120             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1121                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1122                                          &ewtabF,&ewtabFn);
1123             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1124             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1125
1126             /* Analytical LJ-PME */
1127             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1128             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1129             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1130             exponent         = sse41_exp_f(ewcljrsq);
1131             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1132             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1133             /* f6A = 6 * C6grid * (1 - poly) */
1134             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1135             /* f6B = C6grid * exponent * beta^6 */
1136             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1137             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1138             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);
1139
1140             fscal            = _mm_add_ps(felec,fvdw);
1141
1142             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1143
1144             /* Calculate temporary vectorial force */
1145             tx               = _mm_mul_ps(fscal,dx00);
1146             ty               = _mm_mul_ps(fscal,dy00);
1147             tz               = _mm_mul_ps(fscal,dz00);
1148
1149             /* Update vectorial force */
1150             fix0             = _mm_add_ps(fix0,tx);
1151             fiy0             = _mm_add_ps(fiy0,ty);
1152             fiz0             = _mm_add_ps(fiz0,tz);
1153
1154             fjx0             = _mm_add_ps(fjx0,tx);
1155             fjy0             = _mm_add_ps(fjy0,ty);
1156             fjz0             = _mm_add_ps(fjz0,tz);
1157
1158             /**************************
1159              * CALCULATE INTERACTIONS *
1160              **************************/
1161
1162             r10              = _mm_mul_ps(rsq10,rinv10);
1163             r10              = _mm_andnot_ps(dummy_mask,r10);
1164
1165             /* Compute parameters for interactions between i and j atoms */
1166             qq10             = _mm_mul_ps(iq1,jq0);
1167
1168             /* EWALD ELECTROSTATICS */
1169
1170             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1171             ewrt             = _mm_mul_ps(r10,ewtabscale);
1172             ewitab           = _mm_cvttps_epi32(ewrt);
1173             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1174             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1175                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1176                                          &ewtabF,&ewtabFn);
1177             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1178             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1179
1180             fscal            = felec;
1181
1182             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1183
1184             /* Calculate temporary vectorial force */
1185             tx               = _mm_mul_ps(fscal,dx10);
1186             ty               = _mm_mul_ps(fscal,dy10);
1187             tz               = _mm_mul_ps(fscal,dz10);
1188
1189             /* Update vectorial force */
1190             fix1             = _mm_add_ps(fix1,tx);
1191             fiy1             = _mm_add_ps(fiy1,ty);
1192             fiz1             = _mm_add_ps(fiz1,tz);
1193
1194             fjx0             = _mm_add_ps(fjx0,tx);
1195             fjy0             = _mm_add_ps(fjy0,ty);
1196             fjz0             = _mm_add_ps(fjz0,tz);
1197
1198             /**************************
1199              * CALCULATE INTERACTIONS *
1200              **************************/
1201
1202             r20              = _mm_mul_ps(rsq20,rinv20);
1203             r20              = _mm_andnot_ps(dummy_mask,r20);
1204
1205             /* Compute parameters for interactions between i and j atoms */
1206             qq20             = _mm_mul_ps(iq2,jq0);
1207
1208             /* EWALD ELECTROSTATICS */
1209
1210             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1211             ewrt             = _mm_mul_ps(r20,ewtabscale);
1212             ewitab           = _mm_cvttps_epi32(ewrt);
1213             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1214             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1215                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1216                                          &ewtabF,&ewtabFn);
1217             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1218             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1219
1220             fscal            = felec;
1221
1222             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1223
1224             /* Calculate temporary vectorial force */
1225             tx               = _mm_mul_ps(fscal,dx20);
1226             ty               = _mm_mul_ps(fscal,dy20);
1227             tz               = _mm_mul_ps(fscal,dz20);
1228
1229             /* Update vectorial force */
1230             fix2             = _mm_add_ps(fix2,tx);
1231             fiy2             = _mm_add_ps(fiy2,ty);
1232             fiz2             = _mm_add_ps(fiz2,tz);
1233
1234             fjx0             = _mm_add_ps(fjx0,tx);
1235             fjy0             = _mm_add_ps(fjy0,ty);
1236             fjz0             = _mm_add_ps(fjz0,tz);
1237
1238             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1239             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1240             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1241             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1242
1243             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1244
1245             /* Inner loop uses 134 flops */
1246         }
1247
1248         /* End of innermost loop */
1249
1250         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1251                                               f+i_coord_offset,fshift+i_shift_offset);
1252
1253         /* Increment number of inner iterations */
1254         inneriter                  += j_index_end - j_index_start;
1255
1256         /* Outer loop uses 18 flops */
1257     }
1258
1259     /* Increment number of outer iterations */
1260     outeriter        += nri;
1261
1262     /* Update outer/inner flops */
1263
1264     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1265 }