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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse4_1_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128           c6grid_00;
 106     __m128           c6grid_10;
 107     __m128           c6grid_20;
 108     __m128           c6grid_30;
 109     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 110     real             *vdwgridparam;
 111     __m128           one_half  = _mm_set1_ps(0.5);
 112     __m128           minus_one = _mm_set1_ps(-1.0);
 113     __m128i          ewitab;
 114     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 115     real             *ewtab;
 116     __m128           dummy_mask,cutoff_mask;
 117     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 118     __m128           one     = _mm_set1_ps(1.0);
 119     __m128           two     = _mm_set1_ps(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm_set1_ps(fr->epsfac);
 132     charge           = mdatoms->chargeA;
 133     nvdwtype         = fr->ntype;
 134     vdwparam         = fr->nbfp;
 135     vdwtype          = mdatoms->typeA;
 136     vdwgridparam     = fr->ljpme_c6grid;
 137     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 138     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 139     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 140
 141     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 142     ewtab            = fr->ic->tabq_coul_FDV0;
 143     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 144     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 145
 146     /* Setup water-specific parameters */
 147     inr              = nlist->iinr[0];
 148     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 149     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 150     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 151     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153     /* Avoid stupid compiler warnings */
 154     jnrA = jnrB = jnrC = jnrD = 0;
 155     j_coord_offsetA = 0;
 156     j_coord_offsetB = 0;
 157     j_coord_offsetC = 0;
 158     j_coord_offsetD = 0;
 159
 160     outeriter        = 0;
 161     inneriter        = 0;
 162
 163     for(iidx=0;iidx<4*DIM;iidx++)
 164     {
 165         scratch[iidx] = 0.0;
 166     }
 167
 168     /* Start outer loop over neighborlists */
 169     for(iidx=0; iidx<nri; iidx++)
 170     {
 171         /* Load shift vector for this list */
 172         i_shift_offset   = DIM*shiftidx[iidx];
 173
 174         /* Load limits for loop over neighbors */
 175         j_index_start    = jindex[iidx];
 176         j_index_end      = jindex[iidx+1];
 177
 178         /* Get outer coordinate index */
 179         inr              = iinr[iidx];
 180         i_coord_offset   = DIM*inr;
 181
 182         /* Load i particle coords and add shift vector */
 183         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 184                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 185
 186         fix0             = _mm_setzero_ps();
 187         fiy0             = _mm_setzero_ps();
 188         fiz0             = _mm_setzero_ps();
 189         fix1             = _mm_setzero_ps();
 190         fiy1             = _mm_setzero_ps();
 191         fiz1             = _mm_setzero_ps();
 192         fix2             = _mm_setzero_ps();
 193         fiy2             = _mm_setzero_ps();
 194         fiz2             = _mm_setzero_ps();
 195         fix3             = _mm_setzero_ps();
 196         fiy3             = _mm_setzero_ps();
 197         fiz3             = _mm_setzero_ps();
 198
 199         /* Reset potential sums */
 200         velecsum         = _mm_setzero_ps();
 201         vvdwsum          = _mm_setzero_ps();
 202
 203         /* Start inner kernel loop */
 204         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 205         {
 206
 207             /* Get j neighbor index, and coordinate index */
 208             jnrA             = jjnr[jidx];
 209             jnrB             = jjnr[jidx+1];
 210             jnrC             = jjnr[jidx+2];
 211             jnrD             = jjnr[jidx+3];
 212             j_coord_offsetA  = DIM*jnrA;
 213             j_coord_offsetB  = DIM*jnrB;
 214             j_coord_offsetC  = DIM*jnrC;
 215             j_coord_offsetD  = DIM*jnrD;
 216
 217             /* load j atom coordinates */
 218             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 219                                               x+j_coord_offsetC,x+j_coord_offsetD,
 220                                               &jx0,&jy0,&jz0);
 221
 222             /* Calculate displacement vector */
 223             dx00             = _mm_sub_ps(ix0,jx0);
 224             dy00             = _mm_sub_ps(iy0,jy0);
 225             dz00             = _mm_sub_ps(iz0,jz0);
 226             dx10             = _mm_sub_ps(ix1,jx0);
 227             dy10             = _mm_sub_ps(iy1,jy0);
 228             dz10             = _mm_sub_ps(iz1,jz0);
 229             dx20             = _mm_sub_ps(ix2,jx0);
 230             dy20             = _mm_sub_ps(iy2,jy0);
 231             dz20             = _mm_sub_ps(iz2,jz0);
 232             dx30             = _mm_sub_ps(ix3,jx0);
 233             dy30             = _mm_sub_ps(iy3,jy0);
 234             dz30             = _mm_sub_ps(iz3,jz0);
 235
 236             /* Calculate squared distance and things based on it */
 237             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 238             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 239             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 240             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 241
 242             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 243             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 244             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 245             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 246
 247             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 248             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 249             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 250             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 251
 252             /* Load parameters for j particles */
 253             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 254                                                               charge+jnrC+0,charge+jnrD+0);
 255             vdwjidx0A        = 2*vdwtype[jnrA+0];
 256             vdwjidx0B        = 2*vdwtype[jnrB+0];
 257             vdwjidx0C        = 2*vdwtype[jnrC+0];
 258             vdwjidx0D        = 2*vdwtype[jnrD+0];
 259
 260             fjx0             = _mm_setzero_ps();
 261             fjy0             = _mm_setzero_ps();
 262             fjz0             = _mm_setzero_ps();
 263
 264             /**************************
 265              * CALCULATE INTERACTIONS *
 266              **************************/
 267
 268             r00              = _mm_mul_ps(rsq00,rinv00);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 272                                          vdwparam+vdwioffset0+vdwjidx0B,
 273                                          vdwparam+vdwioffset0+vdwjidx0C,
 274                                          vdwparam+vdwioffset0+vdwjidx0D,
 275                                          &c6_00,&c12_00);
 276
 277             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 278                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 279                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 280                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 281
 282             /* Analytical LJ-PME */
 283             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 284             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 285             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 286             exponent         = gmx_simd_exp_r(ewcljrsq);
 287             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 288             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 289             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 290             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 291             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 292             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 293             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 294             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);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 298
 299             fscal            = fvdw;
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm_mul_ps(fscal,dx00);
 303             ty               = _mm_mul_ps(fscal,dy00);
 304             tz               = _mm_mul_ps(fscal,dz00);
 305
 306             /* Update vectorial force */
 307             fix0             = _mm_add_ps(fix0,tx);
 308             fiy0             = _mm_add_ps(fiy0,ty);
 309             fiz0             = _mm_add_ps(fiz0,tz);
 310
 311             fjx0             = _mm_add_ps(fjx0,tx);
 312             fjy0             = _mm_add_ps(fjy0,ty);
 313             fjz0             = _mm_add_ps(fjz0,tz);
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             r10              = _mm_mul_ps(rsq10,rinv10);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm_mul_ps(iq1,jq0);
 323
 324             /* EWALD ELECTROSTATICS */
 325
 326             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 327             ewrt             = _mm_mul_ps(r10,ewtabscale);
 328             ewitab           = _mm_cvttps_epi32(ewrt);
 329             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 330             ewitab           = _mm_slli_epi32(ewitab,2);
 331             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 332             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 333             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 334             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 335             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 336             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 337             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 338             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 339             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_ps(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_ps(fscal,dx10);
 348             ty               = _mm_mul_ps(fscal,dy10);
 349             tz               = _mm_mul_ps(fscal,dz10);
 350
 351             /* Update vectorial force */
 352             fix1             = _mm_add_ps(fix1,tx);
 353             fiy1             = _mm_add_ps(fiy1,ty);
 354             fiz1             = _mm_add_ps(fiz1,tz);
 355
 356             fjx0             = _mm_add_ps(fjx0,tx);
 357             fjy0             = _mm_add_ps(fjy0,ty);
 358             fjz0             = _mm_add_ps(fjz0,tz);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             r20              = _mm_mul_ps(rsq20,rinv20);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq20             = _mm_mul_ps(iq2,jq0);
 368
 369             /* EWALD ELECTROSTATICS */
 370
 371             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 372             ewrt             = _mm_mul_ps(r20,ewtabscale);
 373             ewitab           = _mm_cvttps_epi32(ewrt);
 374             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 375             ewitab           = _mm_slli_epi32(ewitab,2);
 376             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 377             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 378             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 379             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 380             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 381             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 382             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 383             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 384             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velecsum         = _mm_add_ps(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391             /* Calculate temporary vectorial force */
 392             tx               = _mm_mul_ps(fscal,dx20);
 393             ty               = _mm_mul_ps(fscal,dy20);
 394             tz               = _mm_mul_ps(fscal,dz20);
 395
 396             /* Update vectorial force */
 397             fix2             = _mm_add_ps(fix2,tx);
 398             fiy2             = _mm_add_ps(fiy2,ty);
 399             fiz2             = _mm_add_ps(fiz2,tz);
 400
 401             fjx0             = _mm_add_ps(fjx0,tx);
 402             fjy0             = _mm_add_ps(fjy0,ty);
 403             fjz0             = _mm_add_ps(fjz0,tz);
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             r30              = _mm_mul_ps(rsq30,rinv30);
 410
 411             /* Compute parameters for interactions between i and j atoms */
 412             qq30             = _mm_mul_ps(iq3,jq0);
 413
 414             /* EWALD ELECTROSTATICS */
 415
 416             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 417             ewrt             = _mm_mul_ps(r30,ewtabscale);
 418             ewitab           = _mm_cvttps_epi32(ewrt);
 419             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 420             ewitab           = _mm_slli_epi32(ewitab,2);
 421             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 422             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 423             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 424             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 425             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 426             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 427             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 428             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 429             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velecsum         = _mm_add_ps(velecsum,velec);
 433
 434             fscal            = felec;
 435
 436             /* Calculate temporary vectorial force */
 437             tx               = _mm_mul_ps(fscal,dx30);
 438             ty               = _mm_mul_ps(fscal,dy30);
 439             tz               = _mm_mul_ps(fscal,dz30);
 440
 441             /* Update vectorial force */
 442             fix3             = _mm_add_ps(fix3,tx);
 443             fiy3             = _mm_add_ps(fiy3,ty);
 444             fiz3             = _mm_add_ps(fiz3,tz);
 445
 446             fjx0             = _mm_add_ps(fjx0,tx);
 447             fjy0             = _mm_add_ps(fjy0,ty);
 448             fjz0             = _mm_add_ps(fjz0,tz);
 449
 450             fjptrA             = f+j_coord_offsetA;
 451             fjptrB             = f+j_coord_offsetB;
 452             fjptrC             = f+j_coord_offsetC;
 453             fjptrD             = f+j_coord_offsetD;
 454
 455             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 456
 457             /* Inner loop uses 174 flops */
 458         }
 459
 460         if(jidx<j_index_end)
 461         {
 462
 463             /* Get j neighbor index, and coordinate index */
 464             jnrlistA         = jjnr[jidx];
 465             jnrlistB         = jjnr[jidx+1];
 466             jnrlistC         = jjnr[jidx+2];
 467             jnrlistD         = jjnr[jidx+3];
 468             /* Sign of each element will be negative for non-real atoms.
 469              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 470              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 471              */
 472             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 473             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 474             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 475             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 476             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 477             j_coord_offsetA  = DIM*jnrA;
 478             j_coord_offsetB  = DIM*jnrB;
 479             j_coord_offsetC  = DIM*jnrC;
 480             j_coord_offsetD  = DIM*jnrD;
 481
 482             /* load j atom coordinates */
 483             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 484                                               x+j_coord_offsetC,x+j_coord_offsetD,
 485                                               &jx0,&jy0,&jz0);
 486
 487             /* Calculate displacement vector */
 488             dx00             = _mm_sub_ps(ix0,jx0);
 489             dy00             = _mm_sub_ps(iy0,jy0);
 490             dz00             = _mm_sub_ps(iz0,jz0);
 491             dx10             = _mm_sub_ps(ix1,jx0);
 492             dy10             = _mm_sub_ps(iy1,jy0);
 493             dz10             = _mm_sub_ps(iz1,jz0);
 494             dx20             = _mm_sub_ps(ix2,jx0);
 495             dy20             = _mm_sub_ps(iy2,jy0);
 496             dz20             = _mm_sub_ps(iz2,jz0);
 497             dx30             = _mm_sub_ps(ix3,jx0);
 498             dy30             = _mm_sub_ps(iy3,jy0);
 499             dz30             = _mm_sub_ps(iz3,jz0);
 500
 501             /* Calculate squared distance and things based on it */
 502             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 503             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 504             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 505             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 506
 507             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 508             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 509             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 510             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 511
 512             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 513             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 514             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 515             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 516
 517             /* Load parameters for j particles */
 518             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 519                                                               charge+jnrC+0,charge+jnrD+0);
 520             vdwjidx0A        = 2*vdwtype[jnrA+0];
 521             vdwjidx0B        = 2*vdwtype[jnrB+0];
 522             vdwjidx0C        = 2*vdwtype[jnrC+0];
 523             vdwjidx0D        = 2*vdwtype[jnrD+0];
 524
 525             fjx0             = _mm_setzero_ps();
 526             fjy0             = _mm_setzero_ps();
 527             fjz0             = _mm_setzero_ps();
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             r00              = _mm_mul_ps(rsq00,rinv00);
 534             r00              = _mm_andnot_ps(dummy_mask,r00);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 538                                          vdwparam+vdwioffset0+vdwjidx0B,
 539                                          vdwparam+vdwioffset0+vdwjidx0C,
 540                                          vdwparam+vdwioffset0+vdwjidx0D,
 541                                          &c6_00,&c12_00);
 542
 543             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 544                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 545                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 546                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 547
 548             /* Analytical LJ-PME */
 549             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 550             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 551             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 552             exponent         = gmx_simd_exp_r(ewcljrsq);
 553             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 554             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 555             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 556             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 557             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 558             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 559             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 560             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);
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 564             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 565
 566             fscal            = fvdw;
 567
 568             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 569
 570             /* Calculate temporary vectorial force */
 571             tx               = _mm_mul_ps(fscal,dx00);
 572             ty               = _mm_mul_ps(fscal,dy00);
 573             tz               = _mm_mul_ps(fscal,dz00);
 574
 575             /* Update vectorial force */
 576             fix0             = _mm_add_ps(fix0,tx);
 577             fiy0             = _mm_add_ps(fiy0,ty);
 578             fiz0             = _mm_add_ps(fiz0,tz);
 579
 580             fjx0             = _mm_add_ps(fjx0,tx);
 581             fjy0             = _mm_add_ps(fjy0,ty);
 582             fjz0             = _mm_add_ps(fjz0,tz);
 583
 584             /**************************
 585              * CALCULATE INTERACTIONS *
 586              **************************/
 587
 588             r10              = _mm_mul_ps(rsq10,rinv10);
 589             r10              = _mm_andnot_ps(dummy_mask,r10);
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq10             = _mm_mul_ps(iq1,jq0);
 593
 594             /* EWALD ELECTROSTATICS */
 595
 596             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 597             ewrt             = _mm_mul_ps(r10,ewtabscale);
 598             ewitab           = _mm_cvttps_epi32(ewrt);
 599             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 600             ewitab           = _mm_slli_epi32(ewitab,2);
 601             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 602             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 603             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 604             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 605             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 606             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 607             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 608             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 609             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 610
 611             /* Update potential sum for this i atom from the interaction with this j atom. */
 612             velec            = _mm_andnot_ps(dummy_mask,velec);
 613             velecsum         = _mm_add_ps(velecsum,velec);
 614
 615             fscal            = felec;
 616
 617             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm_mul_ps(fscal,dx10);
 621             ty               = _mm_mul_ps(fscal,dy10);
 622             tz               = _mm_mul_ps(fscal,dz10);
 623
 624             /* Update vectorial force */
 625             fix1             = _mm_add_ps(fix1,tx);
 626             fiy1             = _mm_add_ps(fiy1,ty);
 627             fiz1             = _mm_add_ps(fiz1,tz);
 628
 629             fjx0             = _mm_add_ps(fjx0,tx);
 630             fjy0             = _mm_add_ps(fjy0,ty);
 631             fjz0             = _mm_add_ps(fjz0,tz);
 632
 633             /**************************
 634              * CALCULATE INTERACTIONS *
 635              **************************/
 636
 637             r20              = _mm_mul_ps(rsq20,rinv20);
 638             r20              = _mm_andnot_ps(dummy_mask,r20);
 639
 640             /* Compute parameters for interactions between i and j atoms */
 641             qq20             = _mm_mul_ps(iq2,jq0);
 642
 643             /* EWALD ELECTROSTATICS */
 644
 645             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 646             ewrt             = _mm_mul_ps(r20,ewtabscale);
 647             ewitab           = _mm_cvttps_epi32(ewrt);
 648             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 649             ewitab           = _mm_slli_epi32(ewitab,2);
 650             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 651             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 652             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 653             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 654             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 655             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 656             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 657             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 658             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 659
 660             /* Update potential sum for this i atom from the interaction with this j atom. */
 661             velec            = _mm_andnot_ps(dummy_mask,velec);
 662             velecsum         = _mm_add_ps(velecsum,velec);
 663
 664             fscal            = felec;
 665
 666             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm_mul_ps(fscal,dx20);
 670             ty               = _mm_mul_ps(fscal,dy20);
 671             tz               = _mm_mul_ps(fscal,dz20);
 672
 673             /* Update vectorial force */
 674             fix2             = _mm_add_ps(fix2,tx);
 675             fiy2             = _mm_add_ps(fiy2,ty);
 676             fiz2             = _mm_add_ps(fiz2,tz);
 677
 678             fjx0             = _mm_add_ps(fjx0,tx);
 679             fjy0             = _mm_add_ps(fjy0,ty);
 680             fjz0             = _mm_add_ps(fjz0,tz);
 681
 682             /**************************
 683              * CALCULATE INTERACTIONS *
 684              **************************/
 685
 686             r30              = _mm_mul_ps(rsq30,rinv30);
 687             r30              = _mm_andnot_ps(dummy_mask,r30);
 688
 689             /* Compute parameters for interactions between i and j atoms */
 690             qq30             = _mm_mul_ps(iq3,jq0);
 691
 692             /* EWALD ELECTROSTATICS */
 693
 694             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 695             ewrt             = _mm_mul_ps(r30,ewtabscale);
 696             ewitab           = _mm_cvttps_epi32(ewrt);
 697             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 698             ewitab           = _mm_slli_epi32(ewitab,2);
 699             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 700             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 701             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 702             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 703             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 704             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 705             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 706             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 707             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 708
 709             /* Update potential sum for this i atom from the interaction with this j atom. */
 710             velec            = _mm_andnot_ps(dummy_mask,velec);
 711             velecsum         = _mm_add_ps(velecsum,velec);
 712
 713             fscal            = felec;
 714
 715             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 716
 717             /* Calculate temporary vectorial force */
 718             tx               = _mm_mul_ps(fscal,dx30);
 719             ty               = _mm_mul_ps(fscal,dy30);
 720             tz               = _mm_mul_ps(fscal,dz30);
 721
 722             /* Update vectorial force */
 723             fix3             = _mm_add_ps(fix3,tx);
 724             fiy3             = _mm_add_ps(fiy3,ty);
 725             fiz3             = _mm_add_ps(fiz3,tz);
 726
 727             fjx0             = _mm_add_ps(fjx0,tx);
 728             fjy0             = _mm_add_ps(fjy0,ty);
 729             fjz0             = _mm_add_ps(fjz0,tz);
 730
 731             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 732             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 733             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 734             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 735
 736             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 737
 738             /* Inner loop uses 178 flops */
 739         }
 740
 741         /* End of innermost loop */
 742
 743         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 744                                               f+i_coord_offset,fshift+i_shift_offset);
 745
 746         ggid                        = gid[iidx];
 747         /* Update potential energies */
 748         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 749         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 750
 751         /* Increment number of inner iterations */
 752         inneriter                  += j_index_end - j_index_start;
 753
 754         /* Outer loop uses 26 flops */
 755     }
 756
 757     /* Increment number of outer iterations */
 758     outeriter        += nri;
 759
 760     /* Update outer/inner flops */
 761
 762     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*178);
 763 }
 764 /*
 765  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse4_1_single
 766  * Electrostatics interaction: Ewald
 767  * VdW interaction:            LJEwald
 768  * Geometry:                   Water4-Particle
 769  * Calculate force/pot:        Force
 770  */
 771 void
 772 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse4_1_single
 773                     (t_nblist                    * gmx_restrict       nlist,
 774                      rvec                        * gmx_restrict          xx,
 775                      rvec                        * gmx_restrict          ff,
 776                      t_forcerec                  * gmx_restrict          fr,
 777                      t_mdatoms                   * gmx_restrict     mdatoms,
 778                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 779                      t_nrnb                      * gmx_restrict        nrnb)
 780 {
 781     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 782      * just 0 for non-waters.
 783      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 784      * jnr indices corresponding to data put in the four positions in the SIMD register.
 785      */
 786     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 787     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 788     int              jnrA,jnrB,jnrC,jnrD;
 789     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 790     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 791     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 792     real             rcutoff_scalar;
 793     real             *shiftvec,*fshift,*x,*f;
 794     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 795     real             scratch[4*DIM];
 796     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 797     int              vdwioffset0;
 798     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 799     int              vdwioffset1;
 800     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 801     int              vdwioffset2;
 802     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 803     int              vdwioffset3;
 804     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 805     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 806     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 807     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 808     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 809     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 810     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 811     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 812     real             *charge;
 813     int              nvdwtype;
 814     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 815     int              *vdwtype;
 816     real             *vdwparam;
 817     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 818     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 819     __m128           c6grid_00;
 820     __m128           c6grid_10;
 821     __m128           c6grid_20;
 822     __m128           c6grid_30;
 823     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 824     real             *vdwgridparam;
 825     __m128           one_half  = _mm_set1_ps(0.5);
 826     __m128           minus_one = _mm_set1_ps(-1.0);
 827     __m128i          ewitab;
 828     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 829     real             *ewtab;
 830     __m128           dummy_mask,cutoff_mask;
 831     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 832     __m128           one     = _mm_set1_ps(1.0);
 833     __m128           two     = _mm_set1_ps(2.0);
 834     x                = xx[0];
 835     f                = ff[0];
 836
 837     nri              = nlist->nri;
 838     iinr             = nlist->iinr;
 839     jindex           = nlist->jindex;
 840     jjnr             = nlist->jjnr;
 841     shiftidx         = nlist->shift;
 842     gid              = nlist->gid;
 843     shiftvec         = fr->shift_vec[0];
 844     fshift           = fr->fshift[0];
 845     facel            = _mm_set1_ps(fr->epsfac);
 846     charge           = mdatoms->chargeA;
 847     nvdwtype         = fr->ntype;
 848     vdwparam         = fr->nbfp;
 849     vdwtype          = mdatoms->typeA;
 850     vdwgridparam     = fr->ljpme_c6grid;
 851     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 852     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 853     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 854
 855     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 856     ewtab            = fr->ic->tabq_coul_F;
 857     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 858     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 859
 860     /* Setup water-specific parameters */
 861     inr              = nlist->iinr[0];
 862     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 863     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 864     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 865     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 866
 867     /* Avoid stupid compiler warnings */
 868     jnrA = jnrB = jnrC = jnrD = 0;
 869     j_coord_offsetA = 0;
 870     j_coord_offsetB = 0;
 871     j_coord_offsetC = 0;
 872     j_coord_offsetD = 0;
 873
 874     outeriter        = 0;
 875     inneriter        = 0;
 876
 877     for(iidx=0;iidx<4*DIM;iidx++)
 878     {
 879         scratch[iidx] = 0.0;
 880     }
 881
 882     /* Start outer loop over neighborlists */
 883     for(iidx=0; iidx<nri; iidx++)
 884     {
 885         /* Load shift vector for this list */
 886         i_shift_offset   = DIM*shiftidx[iidx];
 887
 888         /* Load limits for loop over neighbors */
 889         j_index_start    = jindex[iidx];
 890         j_index_end      = jindex[iidx+1];
 891
 892         /* Get outer coordinate index */
 893         inr              = iinr[iidx];
 894         i_coord_offset   = DIM*inr;
 895
 896         /* Load i particle coords and add shift vector */
 897         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 898                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 899
 900         fix0             = _mm_setzero_ps();
 901         fiy0             = _mm_setzero_ps();
 902         fiz0             = _mm_setzero_ps();
 903         fix1             = _mm_setzero_ps();
 904         fiy1             = _mm_setzero_ps();
 905         fiz1             = _mm_setzero_ps();
 906         fix2             = _mm_setzero_ps();
 907         fiy2             = _mm_setzero_ps();
 908         fiz2             = _mm_setzero_ps();
 909         fix3             = _mm_setzero_ps();
 910         fiy3             = _mm_setzero_ps();
 911         fiz3             = _mm_setzero_ps();
 912
 913         /* Start inner kernel loop */
 914         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 915         {
 916
 917             /* Get j neighbor index, and coordinate index */
 918             jnrA             = jjnr[jidx];
 919             jnrB             = jjnr[jidx+1];
 920             jnrC             = jjnr[jidx+2];
 921             jnrD             = jjnr[jidx+3];
 922             j_coord_offsetA  = DIM*jnrA;
 923             j_coord_offsetB  = DIM*jnrB;
 924             j_coord_offsetC  = DIM*jnrC;
 925             j_coord_offsetD  = DIM*jnrD;
 926
 927             /* load j atom coordinates */
 928             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 929                                               x+j_coord_offsetC,x+j_coord_offsetD,
 930                                               &jx0,&jy0,&jz0);
 931
 932             /* Calculate displacement vector */
 933             dx00             = _mm_sub_ps(ix0,jx0);
 934             dy00             = _mm_sub_ps(iy0,jy0);
 935             dz00             = _mm_sub_ps(iz0,jz0);
 936             dx10             = _mm_sub_ps(ix1,jx0);
 937             dy10             = _mm_sub_ps(iy1,jy0);
 938             dz10             = _mm_sub_ps(iz1,jz0);
 939             dx20             = _mm_sub_ps(ix2,jx0);
 940             dy20             = _mm_sub_ps(iy2,jy0);
 941             dz20             = _mm_sub_ps(iz2,jz0);
 942             dx30             = _mm_sub_ps(ix3,jx0);
 943             dy30             = _mm_sub_ps(iy3,jy0);
 944             dz30             = _mm_sub_ps(iz3,jz0);
 945
 946             /* Calculate squared distance and things based on it */
 947             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 948             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 949             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 950             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 951
 952             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 953             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 954             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 955             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 956
 957             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 958             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 959             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 960             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 961
 962             /* Load parameters for j particles */
 963             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 964                                                               charge+jnrC+0,charge+jnrD+0);
 965             vdwjidx0A        = 2*vdwtype[jnrA+0];
 966             vdwjidx0B        = 2*vdwtype[jnrB+0];
 967             vdwjidx0C        = 2*vdwtype[jnrC+0];
 968             vdwjidx0D        = 2*vdwtype[jnrD+0];
 969
 970             fjx0             = _mm_setzero_ps();
 971             fjy0             = _mm_setzero_ps();
 972             fjz0             = _mm_setzero_ps();
 973
 974             /**************************
 975              * CALCULATE INTERACTIONS *
 976              **************************/
 977
 978             r00              = _mm_mul_ps(rsq00,rinv00);
 979
 980             /* Compute parameters for interactions between i and j atoms */
 981             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 982                                          vdwparam+vdwioffset0+vdwjidx0B,
 983                                          vdwparam+vdwioffset0+vdwjidx0C,
 984                                          vdwparam+vdwioffset0+vdwjidx0D,
 985                                          &c6_00,&c12_00);
 986
 987             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 988                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 989                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 990                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 991
 992             /* Analytical LJ-PME */
 993             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 994             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 995             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 996             exponent         = gmx_simd_exp_r(ewcljrsq);
 997             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 998             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 999             /* f6A = 6 * C6grid * (1 - poly) */
1000             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1001             /* f6B = C6grid * exponent * beta^6 */
1002             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1003             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1004             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);
1005
1006             fscal            = fvdw;
1007
1008             /* Calculate temporary vectorial force */
1009             tx               = _mm_mul_ps(fscal,dx00);
1010             ty               = _mm_mul_ps(fscal,dy00);
1011             tz               = _mm_mul_ps(fscal,dz00);
1012
1013             /* Update vectorial force */
1014             fix0             = _mm_add_ps(fix0,tx);
1015             fiy0             = _mm_add_ps(fiy0,ty);
1016             fiz0             = _mm_add_ps(fiz0,tz);
1017
1018             fjx0             = _mm_add_ps(fjx0,tx);
1019             fjy0             = _mm_add_ps(fjy0,ty);
1020             fjz0             = _mm_add_ps(fjz0,tz);
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             r10              = _mm_mul_ps(rsq10,rinv10);
1027
1028             /* Compute parameters for interactions between i and j atoms */
1029             qq10             = _mm_mul_ps(iq1,jq0);
1030
1031             /* EWALD ELECTROSTATICS */
1032
1033             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1034             ewrt             = _mm_mul_ps(r10,ewtabscale);
1035             ewitab           = _mm_cvttps_epi32(ewrt);
1036             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1037             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1038                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1039                                          &ewtabF,&ewtabFn);
1040             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1041             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1042
1043             fscal            = felec;
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm_mul_ps(fscal,dx10);
1047             ty               = _mm_mul_ps(fscal,dy10);
1048             tz               = _mm_mul_ps(fscal,dz10);
1049
1050             /* Update vectorial force */
1051             fix1             = _mm_add_ps(fix1,tx);
1052             fiy1             = _mm_add_ps(fiy1,ty);
1053             fiz1             = _mm_add_ps(fiz1,tz);
1054
1055             fjx0             = _mm_add_ps(fjx0,tx);
1056             fjy0             = _mm_add_ps(fjy0,ty);
1057             fjz0             = _mm_add_ps(fjz0,tz);
1058
1059             /**************************
1060              * CALCULATE INTERACTIONS *
1061              **************************/
1062
1063             r20              = _mm_mul_ps(rsq20,rinv20);
1064
1065             /* Compute parameters for interactions between i and j atoms */
1066             qq20             = _mm_mul_ps(iq2,jq0);
1067
1068             /* EWALD ELECTROSTATICS */
1069
1070             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1071             ewrt             = _mm_mul_ps(r20,ewtabscale);
1072             ewitab           = _mm_cvttps_epi32(ewrt);
1073             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1074             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1075                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1076                                          &ewtabF,&ewtabFn);
1077             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1078             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1079
1080             fscal            = felec;
1081
1082             /* Calculate temporary vectorial force */
1083             tx               = _mm_mul_ps(fscal,dx20);
1084             ty               = _mm_mul_ps(fscal,dy20);
1085             tz               = _mm_mul_ps(fscal,dz20);
1086
1087             /* Update vectorial force */
1088             fix2             = _mm_add_ps(fix2,tx);
1089             fiy2             = _mm_add_ps(fiy2,ty);
1090             fiz2             = _mm_add_ps(fiz2,tz);
1091
1092             fjx0             = _mm_add_ps(fjx0,tx);
1093             fjy0             = _mm_add_ps(fjy0,ty);
1094             fjz0             = _mm_add_ps(fjz0,tz);
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             r30              = _mm_mul_ps(rsq30,rinv30);
1101
1102             /* Compute parameters for interactions between i and j atoms */
1103             qq30             = _mm_mul_ps(iq3,jq0);
1104
1105             /* EWALD ELECTROSTATICS */
1106
1107             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1108             ewrt             = _mm_mul_ps(r30,ewtabscale);
1109             ewitab           = _mm_cvttps_epi32(ewrt);
1110             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1111             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1112                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1113                                          &ewtabF,&ewtabFn);
1114             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1115             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1116
1117             fscal            = felec;
1118
1119             /* Calculate temporary vectorial force */
1120             tx               = _mm_mul_ps(fscal,dx30);
1121             ty               = _mm_mul_ps(fscal,dy30);
1122             tz               = _mm_mul_ps(fscal,dz30);
1123
1124             /* Update vectorial force */
1125             fix3             = _mm_add_ps(fix3,tx);
1126             fiy3             = _mm_add_ps(fiy3,ty);
1127             fiz3             = _mm_add_ps(fiz3,tz);
1128
1129             fjx0             = _mm_add_ps(fjx0,tx);
1130             fjy0             = _mm_add_ps(fjy0,ty);
1131             fjz0             = _mm_add_ps(fjz0,tz);
1132
1133             fjptrA             = f+j_coord_offsetA;
1134             fjptrB             = f+j_coord_offsetB;
1135             fjptrC             = f+j_coord_offsetC;
1136             fjptrD             = f+j_coord_offsetD;
1137
1138             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1139
1140             /* Inner loop uses 154 flops */
1141         }
1142
1143         if(jidx<j_index_end)
1144         {
1145
1146             /* Get j neighbor index, and coordinate index */
1147             jnrlistA         = jjnr[jidx];
1148             jnrlistB         = jjnr[jidx+1];
1149             jnrlistC         = jjnr[jidx+2];
1150             jnrlistD         = jjnr[jidx+3];
1151             /* Sign of each element will be negative for non-real atoms.
1152              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1153              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1154              */
1155             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1156             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1157             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1158             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1159             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1160             j_coord_offsetA  = DIM*jnrA;
1161             j_coord_offsetB  = DIM*jnrB;
1162             j_coord_offsetC  = DIM*jnrC;
1163             j_coord_offsetD  = DIM*jnrD;
1164
1165             /* load j atom coordinates */
1166             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1167                                               x+j_coord_offsetC,x+j_coord_offsetD,
1168                                               &jx0,&jy0,&jz0);
1169
1170             /* Calculate displacement vector */
1171             dx00             = _mm_sub_ps(ix0,jx0);
1172             dy00             = _mm_sub_ps(iy0,jy0);
1173             dz00             = _mm_sub_ps(iz0,jz0);
1174             dx10             = _mm_sub_ps(ix1,jx0);
1175             dy10             = _mm_sub_ps(iy1,jy0);
1176             dz10             = _mm_sub_ps(iz1,jz0);
1177             dx20             = _mm_sub_ps(ix2,jx0);
1178             dy20             = _mm_sub_ps(iy2,jy0);
1179             dz20             = _mm_sub_ps(iz2,jz0);
1180             dx30             = _mm_sub_ps(ix3,jx0);
1181             dy30             = _mm_sub_ps(iy3,jy0);
1182             dz30             = _mm_sub_ps(iz3,jz0);
1183
1184             /* Calculate squared distance and things based on it */
1185             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1186             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1187             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1188             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1189
1190             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1191             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1192             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1193             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1194
1195             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1196             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1197             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1198             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1199
1200             /* Load parameters for j particles */
1201             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1202                                                               charge+jnrC+0,charge+jnrD+0);
1203             vdwjidx0A        = 2*vdwtype[jnrA+0];
1204             vdwjidx0B        = 2*vdwtype[jnrB+0];
1205             vdwjidx0C        = 2*vdwtype[jnrC+0];
1206             vdwjidx0D        = 2*vdwtype[jnrD+0];
1207
1208             fjx0             = _mm_setzero_ps();
1209             fjy0             = _mm_setzero_ps();
1210             fjz0             = _mm_setzero_ps();
1211
1212             /**************************
1213              * CALCULATE INTERACTIONS *
1214              **************************/
1215
1216             r00              = _mm_mul_ps(rsq00,rinv00);
1217             r00              = _mm_andnot_ps(dummy_mask,r00);
1218
1219             /* Compute parameters for interactions between i and j atoms */
1220             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1221                                          vdwparam+vdwioffset0+vdwjidx0B,
1222                                          vdwparam+vdwioffset0+vdwjidx0C,
1223                                          vdwparam+vdwioffset0+vdwjidx0D,
1224                                          &c6_00,&c12_00);
1225
1226             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1227                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1228                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1229                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1230
1231             /* Analytical LJ-PME */
1232             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1233             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1234             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1235             exponent         = gmx_simd_exp_r(ewcljrsq);
1236             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1237             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1238             /* f6A = 6 * C6grid * (1 - poly) */
1239             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1240             /* f6B = C6grid * exponent * beta^6 */
1241             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1242             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1243             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);
1244
1245             fscal            = fvdw;
1246
1247             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1248
1249             /* Calculate temporary vectorial force */
1250             tx               = _mm_mul_ps(fscal,dx00);
1251             ty               = _mm_mul_ps(fscal,dy00);
1252             tz               = _mm_mul_ps(fscal,dz00);
1253
1254             /* Update vectorial force */
1255             fix0             = _mm_add_ps(fix0,tx);
1256             fiy0             = _mm_add_ps(fiy0,ty);
1257             fiz0             = _mm_add_ps(fiz0,tz);
1258
1259             fjx0             = _mm_add_ps(fjx0,tx);
1260             fjy0             = _mm_add_ps(fjy0,ty);
1261             fjz0             = _mm_add_ps(fjz0,tz);
1262
1263             /**************************
1264              * CALCULATE INTERACTIONS *
1265              **************************/
1266
1267             r10              = _mm_mul_ps(rsq10,rinv10);
1268             r10              = _mm_andnot_ps(dummy_mask,r10);
1269
1270             /* Compute parameters for interactions between i and j atoms */
1271             qq10             = _mm_mul_ps(iq1,jq0);
1272
1273             /* EWALD ELECTROSTATICS */
1274
1275             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1276             ewrt             = _mm_mul_ps(r10,ewtabscale);
1277             ewitab           = _mm_cvttps_epi32(ewrt);
1278             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1279             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1280                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1281                                          &ewtabF,&ewtabFn);
1282             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1283             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1284
1285             fscal            = felec;
1286
1287             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1288
1289             /* Calculate temporary vectorial force */
1290             tx               = _mm_mul_ps(fscal,dx10);
1291             ty               = _mm_mul_ps(fscal,dy10);
1292             tz               = _mm_mul_ps(fscal,dz10);
1293
1294             /* Update vectorial force */
1295             fix1             = _mm_add_ps(fix1,tx);
1296             fiy1             = _mm_add_ps(fiy1,ty);
1297             fiz1             = _mm_add_ps(fiz1,tz);
1298
1299             fjx0             = _mm_add_ps(fjx0,tx);
1300             fjy0             = _mm_add_ps(fjy0,ty);
1301             fjz0             = _mm_add_ps(fjz0,tz);
1302
1303             /**************************
1304              * CALCULATE INTERACTIONS *
1305              **************************/
1306
1307             r20              = _mm_mul_ps(rsq20,rinv20);
1308             r20              = _mm_andnot_ps(dummy_mask,r20);
1309
1310             /* Compute parameters for interactions between i and j atoms */
1311             qq20             = _mm_mul_ps(iq2,jq0);
1312
1313             /* EWALD ELECTROSTATICS */
1314
1315             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1316             ewrt             = _mm_mul_ps(r20,ewtabscale);
1317             ewitab           = _mm_cvttps_epi32(ewrt);
1318             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1319             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1320                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1321                                          &ewtabF,&ewtabFn);
1322             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1323             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1324
1325             fscal            = felec;
1326
1327             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1328
1329             /* Calculate temporary vectorial force */
1330             tx               = _mm_mul_ps(fscal,dx20);
1331             ty               = _mm_mul_ps(fscal,dy20);
1332             tz               = _mm_mul_ps(fscal,dz20);
1333
1334             /* Update vectorial force */
1335             fix2             = _mm_add_ps(fix2,tx);
1336             fiy2             = _mm_add_ps(fiy2,ty);
1337             fiz2             = _mm_add_ps(fiz2,tz);
1338
1339             fjx0             = _mm_add_ps(fjx0,tx);
1340             fjy0             = _mm_add_ps(fjy0,ty);
1341             fjz0             = _mm_add_ps(fjz0,tz);
1342
1343             /**************************
1344              * CALCULATE INTERACTIONS *
1345              **************************/
1346
1347             r30              = _mm_mul_ps(rsq30,rinv30);
1348             r30              = _mm_andnot_ps(dummy_mask,r30);
1349
1350             /* Compute parameters for interactions between i and j atoms */
1351             qq30             = _mm_mul_ps(iq3,jq0);
1352
1353             /* EWALD ELECTROSTATICS */
1354
1355             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1356             ewrt             = _mm_mul_ps(r30,ewtabscale);
1357             ewitab           = _mm_cvttps_epi32(ewrt);
1358             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1359             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1360                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1361                                          &ewtabF,&ewtabFn);
1362             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1363             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1364
1365             fscal            = felec;
1366
1367             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1368
1369             /* Calculate temporary vectorial force */
1370             tx               = _mm_mul_ps(fscal,dx30);
1371             ty               = _mm_mul_ps(fscal,dy30);
1372             tz               = _mm_mul_ps(fscal,dz30);
1373
1374             /* Update vectorial force */
1375             fix3             = _mm_add_ps(fix3,tx);
1376             fiy3             = _mm_add_ps(fiy3,ty);
1377             fiz3             = _mm_add_ps(fiz3,tz);
1378
1379             fjx0             = _mm_add_ps(fjx0,tx);
1380             fjy0             = _mm_add_ps(fjy0,ty);
1381             fjz0             = _mm_add_ps(fjz0,tz);
1382
1383             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1384             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1385             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1386             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1387
1388             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1389
1390             /* Inner loop uses 158 flops */
1391         }
1392
1393         /* End of innermost loop */
1394
1395         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1396                                               f+i_coord_offset,fshift+i_shift_offset);
1397
1398         /* Increment number of inner iterations */
1399         inneriter                  += j_index_end - j_index_start;
1400
1401         /* Outer loop uses 24 flops */
1402     }
1403
1404     /* Increment number of outer iterations */
1405     outeriter        += nri;
1406
1407     /* Update outer/inner flops */
1408
1409     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*158);
1410 }