src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwLJSh_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128i          ewitab;
 108     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 109     real             *ewtab;
 110     __m128           dummy_mask,cutoff_mask;
 111     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 112     __m128           one     = _mm_set1_ps(1.0);
 113     __m128           two     = _mm_set1_ps(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm_set1_ps(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 132     ewtab            = fr->ic->tabq_coul_FDV0;
 133     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 134     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 135
 136     /* Setup water-specific parameters */
 137     inr              = nlist->iinr[0];
 138     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 139     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 140     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 141     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 142
 143     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 144     rcutoff_scalar   = fr->rcoulomb;
 145     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 146     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 147
 148     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 149     rvdw             = _mm_set1_ps(fr->rvdw);
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = jnrC = jnrD = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155     j_coord_offsetC = 0;
 156     j_coord_offsetD = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     for(iidx=0;iidx<4*DIM;iidx++)
 162     {
 163         scratch[iidx] = 0.0;
 164     }
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 183
 184         fix0             = _mm_setzero_ps();
 185         fiy0             = _mm_setzero_ps();
 186         fiz0             = _mm_setzero_ps();
 187         fix1             = _mm_setzero_ps();
 188         fiy1             = _mm_setzero_ps();
 189         fiz1             = _mm_setzero_ps();
 190         fix2             = _mm_setzero_ps();
 191         fiy2             = _mm_setzero_ps();
 192         fiz2             = _mm_setzero_ps();
 193         fix3             = _mm_setzero_ps();
 194         fiy3             = _mm_setzero_ps();
 195         fiz3             = _mm_setzero_ps();
 196
 197         /* Reset potential sums */
 198         velecsum         = _mm_setzero_ps();
 199         vvdwsum          = _mm_setzero_ps();
 200
 201         /* Start inner kernel loop */
 202         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 203         {
 204
 205             /* Get j neighbor index, and coordinate index */
 206             jnrA             = jjnr[jidx];
 207             jnrB             = jjnr[jidx+1];
 208             jnrC             = jjnr[jidx+2];
 209             jnrD             = jjnr[jidx+3];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212             j_coord_offsetC  = DIM*jnrC;
 213             j_coord_offsetD  = DIM*jnrD;
 214
 215             /* load j atom coordinates */
 216             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 217                                               x+j_coord_offsetC,x+j_coord_offsetD,
 218                                               &jx0,&jy0,&jz0);
 219
 220             /* Calculate displacement vector */
 221             dx00             = _mm_sub_ps(ix0,jx0);
 222             dy00             = _mm_sub_ps(iy0,jy0);
 223             dz00             = _mm_sub_ps(iz0,jz0);
 224             dx10             = _mm_sub_ps(ix1,jx0);
 225             dy10             = _mm_sub_ps(iy1,jy0);
 226             dz10             = _mm_sub_ps(iz1,jz0);
 227             dx20             = _mm_sub_ps(ix2,jx0);
 228             dy20             = _mm_sub_ps(iy2,jy0);
 229             dz20             = _mm_sub_ps(iz2,jz0);
 230             dx30             = _mm_sub_ps(ix3,jx0);
 231             dy30             = _mm_sub_ps(iy3,jy0);
 232             dz30             = _mm_sub_ps(iz3,jz0);
 233
 234             /* Calculate squared distance and things based on it */
 235             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 236             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 237             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 238             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 239
 240             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 241             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 242             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 243
 244             rinvsq00         = gmx_mm_inv_ps(rsq00);
 245             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 246             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 247             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 248
 249             /* Load parameters for j particles */
 250             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 251                                                               charge+jnrC+0,charge+jnrD+0);
 252             vdwjidx0A        = 2*vdwtype[jnrA+0];
 253             vdwjidx0B        = 2*vdwtype[jnrB+0];
 254             vdwjidx0C        = 2*vdwtype[jnrC+0];
 255             vdwjidx0D        = 2*vdwtype[jnrD+0];
 256
 257             fjx0             = _mm_setzero_ps();
 258             fjy0             = _mm_setzero_ps();
 259             fjz0             = _mm_setzero_ps();
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             if (gmx_mm_any_lt(rsq00,rcutoff2))
 266             {
 267
 268             /* Compute parameters for interactions between i and j atoms */
 269             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 270                                          vdwparam+vdwioffset0+vdwjidx0B,
 271                                          vdwparam+vdwioffset0+vdwjidx0C,
 272                                          vdwparam+vdwioffset0+vdwjidx0D,
 273                                          &c6_00,&c12_00);
 274
 275             /* LENNARD-JONES DISPERSION/REPULSION */
 276
 277             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 278             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 279             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 280             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 281                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 282             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 283
 284             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 288             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 289
 290             fscal            = fvdw;
 291
 292             fscal            = _mm_and_ps(fscal,cutoff_mask);
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm_mul_ps(fscal,dx00);
 296             ty               = _mm_mul_ps(fscal,dy00);
 297             tz               = _mm_mul_ps(fscal,dz00);
 298
 299             /* Update vectorial force */
 300             fix0             = _mm_add_ps(fix0,tx);
 301             fiy0             = _mm_add_ps(fiy0,ty);
 302             fiz0             = _mm_add_ps(fiz0,tz);
 303
 304             fjx0             = _mm_add_ps(fjx0,tx);
 305             fjy0             = _mm_add_ps(fjy0,ty);
 306             fjz0             = _mm_add_ps(fjz0,tz);
 307
 308             }
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             if (gmx_mm_any_lt(rsq10,rcutoff2))
 315             {
 316
 317             r10              = _mm_mul_ps(rsq10,rinv10);
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm_mul_ps(iq1,jq0);
 321
 322             /* EWALD ELECTROSTATICS */
 323
 324             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 325             ewrt             = _mm_mul_ps(r10,ewtabscale);
 326             ewitab           = _mm_cvttps_epi32(ewrt);
 327             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 328             ewitab           = _mm_slli_epi32(ewitab,2);
 329             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 330             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 331             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 332             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 333             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 334             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 335             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 336             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 337             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 338
 339             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velec            = _mm_and_ps(velec,cutoff_mask);
 343             velecsum         = _mm_add_ps(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             fscal            = _mm_and_ps(fscal,cutoff_mask);
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm_mul_ps(fscal,dx10);
 351             ty               = _mm_mul_ps(fscal,dy10);
 352             tz               = _mm_mul_ps(fscal,dz10);
 353
 354             /* Update vectorial force */
 355             fix1             = _mm_add_ps(fix1,tx);
 356             fiy1             = _mm_add_ps(fiy1,ty);
 357             fiz1             = _mm_add_ps(fiz1,tz);
 358
 359             fjx0             = _mm_add_ps(fjx0,tx);
 360             fjy0             = _mm_add_ps(fjy0,ty);
 361             fjz0             = _mm_add_ps(fjz0,tz);
 362
 363             }
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             if (gmx_mm_any_lt(rsq20,rcutoff2))
 370             {
 371
 372             r20              = _mm_mul_ps(rsq20,rinv20);
 373
 374             /* Compute parameters for interactions between i and j atoms */
 375             qq20             = _mm_mul_ps(iq2,jq0);
 376
 377             /* EWALD ELECTROSTATICS */
 378
 379             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 380             ewrt             = _mm_mul_ps(r20,ewtabscale);
 381             ewitab           = _mm_cvttps_epi32(ewrt);
 382             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 383             ewitab           = _mm_slli_epi32(ewitab,2);
 384             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 385             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 386             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 387             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 388             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 389             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 390             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 391             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 392             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 393
 394             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 395
 396             /* Update potential sum for this i atom from the interaction with this j atom. */
 397             velec            = _mm_and_ps(velec,cutoff_mask);
 398             velecsum         = _mm_add_ps(velecsum,velec);
 399
 400             fscal            = felec;
 401
 402             fscal            = _mm_and_ps(fscal,cutoff_mask);
 403
 404             /* Calculate temporary vectorial force */
 405             tx               = _mm_mul_ps(fscal,dx20);
 406             ty               = _mm_mul_ps(fscal,dy20);
 407             tz               = _mm_mul_ps(fscal,dz20);
 408
 409             /* Update vectorial force */
 410             fix2             = _mm_add_ps(fix2,tx);
 411             fiy2             = _mm_add_ps(fiy2,ty);
 412             fiz2             = _mm_add_ps(fiz2,tz);
 413
 414             fjx0             = _mm_add_ps(fjx0,tx);
 415             fjy0             = _mm_add_ps(fjy0,ty);
 416             fjz0             = _mm_add_ps(fjz0,tz);
 417
 418             }
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             if (gmx_mm_any_lt(rsq30,rcutoff2))
 425             {
 426
 427             r30              = _mm_mul_ps(rsq30,rinv30);
 428
 429             /* Compute parameters for interactions between i and j atoms */
 430             qq30             = _mm_mul_ps(iq3,jq0);
 431
 432             /* EWALD ELECTROSTATICS */
 433
 434             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 435             ewrt             = _mm_mul_ps(r30,ewtabscale);
 436             ewitab           = _mm_cvttps_epi32(ewrt);
 437             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 438             ewitab           = _mm_slli_epi32(ewitab,2);
 439             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 440             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 441             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 442             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 443             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 444             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 445             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 446             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 447             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 448
 449             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 450
 451             /* Update potential sum for this i atom from the interaction with this j atom. */
 452             velec            = _mm_and_ps(velec,cutoff_mask);
 453             velecsum         = _mm_add_ps(velecsum,velec);
 454
 455             fscal            = felec;
 456
 457             fscal            = _mm_and_ps(fscal,cutoff_mask);
 458
 459             /* Calculate temporary vectorial force */
 460             tx               = _mm_mul_ps(fscal,dx30);
 461             ty               = _mm_mul_ps(fscal,dy30);
 462             tz               = _mm_mul_ps(fscal,dz30);
 463
 464             /* Update vectorial force */
 465             fix3             = _mm_add_ps(fix3,tx);
 466             fiy3             = _mm_add_ps(fiy3,ty);
 467             fiz3             = _mm_add_ps(fiz3,tz);
 468
 469             fjx0             = _mm_add_ps(fjx0,tx);
 470             fjy0             = _mm_add_ps(fjy0,ty);
 471             fjz0             = _mm_add_ps(fjz0,tz);
 472
 473             }
 474
 475             fjptrA             = f+j_coord_offsetA;
 476             fjptrB             = f+j_coord_offsetB;
 477             fjptrC             = f+j_coord_offsetC;
 478             fjptrD             = f+j_coord_offsetD;
 479
 480             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 481
 482             /* Inner loop uses 179 flops */
 483         }
 484
 485         if(jidx<j_index_end)
 486         {
 487
 488             /* Get j neighbor index, and coordinate index */
 489             jnrlistA         = jjnr[jidx];
 490             jnrlistB         = jjnr[jidx+1];
 491             jnrlistC         = jjnr[jidx+2];
 492             jnrlistD         = jjnr[jidx+3];
 493             /* Sign of each element will be negative for non-real atoms.
 494              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 495              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 496              */
 497             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 498             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 499             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 500             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 501             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 502             j_coord_offsetA  = DIM*jnrA;
 503             j_coord_offsetB  = DIM*jnrB;
 504             j_coord_offsetC  = DIM*jnrC;
 505             j_coord_offsetD  = DIM*jnrD;
 506
 507             /* load j atom coordinates */
 508             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 509                                               x+j_coord_offsetC,x+j_coord_offsetD,
 510                                               &jx0,&jy0,&jz0);
 511
 512             /* Calculate displacement vector */
 513             dx00             = _mm_sub_ps(ix0,jx0);
 514             dy00             = _mm_sub_ps(iy0,jy0);
 515             dz00             = _mm_sub_ps(iz0,jz0);
 516             dx10             = _mm_sub_ps(ix1,jx0);
 517             dy10             = _mm_sub_ps(iy1,jy0);
 518             dz10             = _mm_sub_ps(iz1,jz0);
 519             dx20             = _mm_sub_ps(ix2,jx0);
 520             dy20             = _mm_sub_ps(iy2,jy0);
 521             dz20             = _mm_sub_ps(iz2,jz0);
 522             dx30             = _mm_sub_ps(ix3,jx0);
 523             dy30             = _mm_sub_ps(iy3,jy0);
 524             dz30             = _mm_sub_ps(iz3,jz0);
 525
 526             /* Calculate squared distance and things based on it */
 527             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 528             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 529             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 530             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 531
 532             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 533             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 534             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 535
 536             rinvsq00         = gmx_mm_inv_ps(rsq00);
 537             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 538             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 539             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 540
 541             /* Load parameters for j particles */
 542             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 543                                                               charge+jnrC+0,charge+jnrD+0);
 544             vdwjidx0A        = 2*vdwtype[jnrA+0];
 545             vdwjidx0B        = 2*vdwtype[jnrB+0];
 546             vdwjidx0C        = 2*vdwtype[jnrC+0];
 547             vdwjidx0D        = 2*vdwtype[jnrD+0];
 548
 549             fjx0             = _mm_setzero_ps();
 550             fjy0             = _mm_setzero_ps();
 551             fjz0             = _mm_setzero_ps();
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             if (gmx_mm_any_lt(rsq00,rcutoff2))
 558             {
 559
 560             /* Compute parameters for interactions between i and j atoms */
 561             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 562                                          vdwparam+vdwioffset0+vdwjidx0B,
 563                                          vdwparam+vdwioffset0+vdwjidx0C,
 564                                          vdwparam+vdwioffset0+vdwjidx0D,
 565                                          &c6_00,&c12_00);
 566
 567             /* LENNARD-JONES DISPERSION/REPULSION */
 568
 569             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 570             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 571             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 572             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 573                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 574             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 575
 576             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 577
 578             /* Update potential sum for this i atom from the interaction with this j atom. */
 579             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 580             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 581             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 582
 583             fscal            = fvdw;
 584
 585             fscal            = _mm_and_ps(fscal,cutoff_mask);
 586
 587             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 588
 589             /* Calculate temporary vectorial force */
 590             tx               = _mm_mul_ps(fscal,dx00);
 591             ty               = _mm_mul_ps(fscal,dy00);
 592             tz               = _mm_mul_ps(fscal,dz00);
 593
 594             /* Update vectorial force */
 595             fix0             = _mm_add_ps(fix0,tx);
 596             fiy0             = _mm_add_ps(fiy0,ty);
 597             fiz0             = _mm_add_ps(fiz0,tz);
 598
 599             fjx0             = _mm_add_ps(fjx0,tx);
 600             fjy0             = _mm_add_ps(fjy0,ty);
 601             fjz0             = _mm_add_ps(fjz0,tz);
 602
 603             }
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             if (gmx_mm_any_lt(rsq10,rcutoff2))
 610             {
 611
 612             r10              = _mm_mul_ps(rsq10,rinv10);
 613             r10              = _mm_andnot_ps(dummy_mask,r10);
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq10             = _mm_mul_ps(iq1,jq0);
 617
 618             /* EWALD ELECTROSTATICS */
 619
 620             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 621             ewrt             = _mm_mul_ps(r10,ewtabscale);
 622             ewitab           = _mm_cvttps_epi32(ewrt);
 623             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 624             ewitab           = _mm_slli_epi32(ewitab,2);
 625             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 626             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 627             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 628             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 629             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 630             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 631             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 632             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 633             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 634
 635             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 636
 637             /* Update potential sum for this i atom from the interaction with this j atom. */
 638             velec            = _mm_and_ps(velec,cutoff_mask);
 639             velec            = _mm_andnot_ps(dummy_mask,velec);
 640             velecsum         = _mm_add_ps(velecsum,velec);
 641
 642             fscal            = felec;
 643
 644             fscal            = _mm_and_ps(fscal,cutoff_mask);
 645
 646             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm_mul_ps(fscal,dx10);
 650             ty               = _mm_mul_ps(fscal,dy10);
 651             tz               = _mm_mul_ps(fscal,dz10);
 652
 653             /* Update vectorial force */
 654             fix1             = _mm_add_ps(fix1,tx);
 655             fiy1             = _mm_add_ps(fiy1,ty);
 656             fiz1             = _mm_add_ps(fiz1,tz);
 657
 658             fjx0             = _mm_add_ps(fjx0,tx);
 659             fjy0             = _mm_add_ps(fjy0,ty);
 660             fjz0             = _mm_add_ps(fjz0,tz);
 661
 662             }
 663
 664             /**************************
 665              * CALCULATE INTERACTIONS *
 666              **************************/
 667
 668             if (gmx_mm_any_lt(rsq20,rcutoff2))
 669             {
 670
 671             r20              = _mm_mul_ps(rsq20,rinv20);
 672             r20              = _mm_andnot_ps(dummy_mask,r20);
 673
 674             /* Compute parameters for interactions between i and j atoms */
 675             qq20             = _mm_mul_ps(iq2,jq0);
 676
 677             /* EWALD ELECTROSTATICS */
 678
 679             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 680             ewrt             = _mm_mul_ps(r20,ewtabscale);
 681             ewitab           = _mm_cvttps_epi32(ewrt);
 682             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 683             ewitab           = _mm_slli_epi32(ewitab,2);
 684             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 685             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 686             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 687             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 688             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 689             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 690             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 691             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 692             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 693
 694             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 695
 696             /* Update potential sum for this i atom from the interaction with this j atom. */
 697             velec            = _mm_and_ps(velec,cutoff_mask);
 698             velec            = _mm_andnot_ps(dummy_mask,velec);
 699             velecsum         = _mm_add_ps(velecsum,velec);
 700
 701             fscal            = felec;
 702
 703             fscal            = _mm_and_ps(fscal,cutoff_mask);
 704
 705             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 706
 707             /* Calculate temporary vectorial force */
 708             tx               = _mm_mul_ps(fscal,dx20);
 709             ty               = _mm_mul_ps(fscal,dy20);
 710             tz               = _mm_mul_ps(fscal,dz20);
 711
 712             /* Update vectorial force */
 713             fix2             = _mm_add_ps(fix2,tx);
 714             fiy2             = _mm_add_ps(fiy2,ty);
 715             fiz2             = _mm_add_ps(fiz2,tz);
 716
 717             fjx0             = _mm_add_ps(fjx0,tx);
 718             fjy0             = _mm_add_ps(fjy0,ty);
 719             fjz0             = _mm_add_ps(fjz0,tz);
 720
 721             }
 722
 723             /**************************
 724              * CALCULATE INTERACTIONS *
 725              **************************/
 726
 727             if (gmx_mm_any_lt(rsq30,rcutoff2))
 728             {
 729
 730             r30              = _mm_mul_ps(rsq30,rinv30);
 731             r30              = _mm_andnot_ps(dummy_mask,r30);
 732
 733             /* Compute parameters for interactions between i and j atoms */
 734             qq30             = _mm_mul_ps(iq3,jq0);
 735
 736             /* EWALD ELECTROSTATICS */
 737
 738             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 739             ewrt             = _mm_mul_ps(r30,ewtabscale);
 740             ewitab           = _mm_cvttps_epi32(ewrt);
 741             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 742             ewitab           = _mm_slli_epi32(ewitab,2);
 743             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 744             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 745             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 746             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 747             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 748             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 749             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 750             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 751             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 752
 753             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 754
 755             /* Update potential sum for this i atom from the interaction with this j atom. */
 756             velec            = _mm_and_ps(velec,cutoff_mask);
 757             velec            = _mm_andnot_ps(dummy_mask,velec);
 758             velecsum         = _mm_add_ps(velecsum,velec);
 759
 760             fscal            = felec;
 761
 762             fscal            = _mm_and_ps(fscal,cutoff_mask);
 763
 764             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 765
 766             /* Calculate temporary vectorial force */
 767             tx               = _mm_mul_ps(fscal,dx30);
 768             ty               = _mm_mul_ps(fscal,dy30);
 769             tz               = _mm_mul_ps(fscal,dz30);
 770
 771             /* Update vectorial force */
 772             fix3             = _mm_add_ps(fix3,tx);
 773             fiy3             = _mm_add_ps(fiy3,ty);
 774             fiz3             = _mm_add_ps(fiz3,tz);
 775
 776             fjx0             = _mm_add_ps(fjx0,tx);
 777             fjy0             = _mm_add_ps(fjy0,ty);
 778             fjz0             = _mm_add_ps(fjz0,tz);
 779
 780             }
 781
 782             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 783             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 784             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 785             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 786
 787             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 788
 789             /* Inner loop uses 182 flops */
 790         }
 791
 792         /* End of innermost loop */
 793
 794         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 795                                               f+i_coord_offset,fshift+i_shift_offset);
 796
 797         ggid                        = gid[iidx];
 798         /* Update potential energies */
 799         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 800         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 801
 802         /* Increment number of inner iterations */
 803         inneriter                  += j_index_end - j_index_start;
 804
 805         /* Outer loop uses 26 flops */
 806     }
 807
 808     /* Increment number of outer iterations */
 809     outeriter        += nri;
 810
 811     /* Update outer/inner flops */
 812
 813     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*182);
 814 }
 815 /*
 816  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse4_1_single
 817  * Electrostatics interaction: Ewald
 818  * VdW interaction:            LennardJones
 819  * Geometry:                   Water4-Particle
 820  * Calculate force/pot:        Force
 821  */
 822 void
 823 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse4_1_single
 824                     (t_nblist                    * gmx_restrict       nlist,
 825                      rvec                        * gmx_restrict          xx,
 826                      rvec                        * gmx_restrict          ff,
 827                      t_forcerec                  * gmx_restrict          fr,
 828                      t_mdatoms                   * gmx_restrict     mdatoms,
 829                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 830                      t_nrnb                      * gmx_restrict        nrnb)
 831 {
 832     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 833      * just 0 for non-waters.
 834      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 835      * jnr indices corresponding to data put in the four positions in the SIMD register.
 836      */
 837     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 838     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 839     int              jnrA,jnrB,jnrC,jnrD;
 840     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 841     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 842     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 843     real             rcutoff_scalar;
 844     real             *shiftvec,*fshift,*x,*f;
 845     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 846     real             scratch[4*DIM];
 847     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 848     int              vdwioffset0;
 849     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 850     int              vdwioffset1;
 851     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 852     int              vdwioffset2;
 853     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 854     int              vdwioffset3;
 855     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 856     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 857     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 858     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 859     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 860     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 861     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 862     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 863     real             *charge;
 864     int              nvdwtype;
 865     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 866     int              *vdwtype;
 867     real             *vdwparam;
 868     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 869     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 870     __m128i          ewitab;
 871     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 872     real             *ewtab;
 873     __m128           dummy_mask,cutoff_mask;
 874     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 875     __m128           one     = _mm_set1_ps(1.0);
 876     __m128           two     = _mm_set1_ps(2.0);
 877     x                = xx[0];
 878     f                = ff[0];
 879
 880     nri              = nlist->nri;
 881     iinr             = nlist->iinr;
 882     jindex           = nlist->jindex;
 883     jjnr             = nlist->jjnr;
 884     shiftidx         = nlist->shift;
 885     gid              = nlist->gid;
 886     shiftvec         = fr->shift_vec[0];
 887     fshift           = fr->fshift[0];
 888     facel            = _mm_set1_ps(fr->epsfac);
 889     charge           = mdatoms->chargeA;
 890     nvdwtype         = fr->ntype;
 891     vdwparam         = fr->nbfp;
 892     vdwtype          = mdatoms->typeA;
 893
 894     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 895     ewtab            = fr->ic->tabq_coul_F;
 896     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 897     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 898
 899     /* Setup water-specific parameters */
 900     inr              = nlist->iinr[0];
 901     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 902     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 903     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 904     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 905
 906     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 907     rcutoff_scalar   = fr->rcoulomb;
 908     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 909     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 910
 911     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 912     rvdw             = _mm_set1_ps(fr->rvdw);
 913
 914     /* Avoid stupid compiler warnings */
 915     jnrA = jnrB = jnrC = jnrD = 0;
 916     j_coord_offsetA = 0;
 917     j_coord_offsetB = 0;
 918     j_coord_offsetC = 0;
 919     j_coord_offsetD = 0;
 920
 921     outeriter        = 0;
 922     inneriter        = 0;
 923
 924     for(iidx=0;iidx<4*DIM;iidx++)
 925     {
 926         scratch[iidx] = 0.0;
 927     }
 928
 929     /* Start outer loop over neighborlists */
 930     for(iidx=0; iidx<nri; iidx++)
 931     {
 932         /* Load shift vector for this list */
 933         i_shift_offset   = DIM*shiftidx[iidx];
 934
 935         /* Load limits for loop over neighbors */
 936         j_index_start    = jindex[iidx];
 937         j_index_end      = jindex[iidx+1];
 938
 939         /* Get outer coordinate index */
 940         inr              = iinr[iidx];
 941         i_coord_offset   = DIM*inr;
 942
 943         /* Load i particle coords and add shift vector */
 944         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 945                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 946
 947         fix0             = _mm_setzero_ps();
 948         fiy0             = _mm_setzero_ps();
 949         fiz0             = _mm_setzero_ps();
 950         fix1             = _mm_setzero_ps();
 951         fiy1             = _mm_setzero_ps();
 952         fiz1             = _mm_setzero_ps();
 953         fix2             = _mm_setzero_ps();
 954         fiy2             = _mm_setzero_ps();
 955         fiz2             = _mm_setzero_ps();
 956         fix3             = _mm_setzero_ps();
 957         fiy3             = _mm_setzero_ps();
 958         fiz3             = _mm_setzero_ps();
 959
 960         /* Start inner kernel loop */
 961         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 962         {
 963
 964             /* Get j neighbor index, and coordinate index */
 965             jnrA             = jjnr[jidx];
 966             jnrB             = jjnr[jidx+1];
 967             jnrC             = jjnr[jidx+2];
 968             jnrD             = jjnr[jidx+3];
 969             j_coord_offsetA  = DIM*jnrA;
 970             j_coord_offsetB  = DIM*jnrB;
 971             j_coord_offsetC  = DIM*jnrC;
 972             j_coord_offsetD  = DIM*jnrD;
 973
 974             /* load j atom coordinates */
 975             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 976                                               x+j_coord_offsetC,x+j_coord_offsetD,
 977                                               &jx0,&jy0,&jz0);
 978
 979             /* Calculate displacement vector */
 980             dx00             = _mm_sub_ps(ix0,jx0);
 981             dy00             = _mm_sub_ps(iy0,jy0);
 982             dz00             = _mm_sub_ps(iz0,jz0);
 983             dx10             = _mm_sub_ps(ix1,jx0);
 984             dy10             = _mm_sub_ps(iy1,jy0);
 985             dz10             = _mm_sub_ps(iz1,jz0);
 986             dx20             = _mm_sub_ps(ix2,jx0);
 987             dy20             = _mm_sub_ps(iy2,jy0);
 988             dz20             = _mm_sub_ps(iz2,jz0);
 989             dx30             = _mm_sub_ps(ix3,jx0);
 990             dy30             = _mm_sub_ps(iy3,jy0);
 991             dz30             = _mm_sub_ps(iz3,jz0);
 992
 993             /* Calculate squared distance and things based on it */
 994             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 995             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 996             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 997             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 998
 999             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1000             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1001             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1002
1003             rinvsq00         = gmx_mm_inv_ps(rsq00);
1004             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1005             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1006             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1007
1008             /* Load parameters for j particles */
1009             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1010                                                               charge+jnrC+0,charge+jnrD+0);
1011             vdwjidx0A        = 2*vdwtype[jnrA+0];
1012             vdwjidx0B        = 2*vdwtype[jnrB+0];
1013             vdwjidx0C        = 2*vdwtype[jnrC+0];
1014             vdwjidx0D        = 2*vdwtype[jnrD+0];
1015
1016             fjx0             = _mm_setzero_ps();
1017             fjy0             = _mm_setzero_ps();
1018             fjz0             = _mm_setzero_ps();
1019
1020             /**************************
1021              * CALCULATE INTERACTIONS *
1022              **************************/
1023
1024             if (gmx_mm_any_lt(rsq00,rcutoff2))
1025             {
1026
1027             /* Compute parameters for interactions between i and j atoms */
1028             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1029                                          vdwparam+vdwioffset0+vdwjidx0B,
1030                                          vdwparam+vdwioffset0+vdwjidx0C,
1031                                          vdwparam+vdwioffset0+vdwjidx0D,
1032                                          &c6_00,&c12_00);
1033
1034             /* LENNARD-JONES DISPERSION/REPULSION */
1035
1036             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1037             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1038
1039             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1040
1041             fscal            = fvdw;
1042
1043             fscal            = _mm_and_ps(fscal,cutoff_mask);
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm_mul_ps(fscal,dx00);
1047             ty               = _mm_mul_ps(fscal,dy00);
1048             tz               = _mm_mul_ps(fscal,dz00);
1049
1050             /* Update vectorial force */
1051             fix0             = _mm_add_ps(fix0,tx);
1052             fiy0             = _mm_add_ps(fiy0,ty);
1053             fiz0             = _mm_add_ps(fiz0,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
1061             /**************************
1062              * CALCULATE INTERACTIONS *
1063              **************************/
1064
1065             if (gmx_mm_any_lt(rsq10,rcutoff2))
1066             {
1067
1068             r10              = _mm_mul_ps(rsq10,rinv10);
1069
1070             /* Compute parameters for interactions between i and j atoms */
1071             qq10             = _mm_mul_ps(iq1,jq0);
1072
1073             /* EWALD ELECTROSTATICS */
1074
1075             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1076             ewrt             = _mm_mul_ps(r10,ewtabscale);
1077             ewitab           = _mm_cvttps_epi32(ewrt);
1078             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1079             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1080                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1081                                          &ewtabF,&ewtabFn);
1082             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1083             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1084
1085             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1086
1087             fscal            = felec;
1088
1089             fscal            = _mm_and_ps(fscal,cutoff_mask);
1090
1091             /* Calculate temporary vectorial force */
1092             tx               = _mm_mul_ps(fscal,dx10);
1093             ty               = _mm_mul_ps(fscal,dy10);
1094             tz               = _mm_mul_ps(fscal,dz10);
1095
1096             /* Update vectorial force */
1097             fix1             = _mm_add_ps(fix1,tx);
1098             fiy1             = _mm_add_ps(fiy1,ty);
1099             fiz1             = _mm_add_ps(fiz1,tz);
1100
1101             fjx0             = _mm_add_ps(fjx0,tx);
1102             fjy0             = _mm_add_ps(fjy0,ty);
1103             fjz0             = _mm_add_ps(fjz0,tz);
1104
1105             }
1106
1107             /**************************
1108              * CALCULATE INTERACTIONS *
1109              **************************/
1110
1111             if (gmx_mm_any_lt(rsq20,rcutoff2))
1112             {
1113
1114             r20              = _mm_mul_ps(rsq20,rinv20);
1115
1116             /* Compute parameters for interactions between i and j atoms */
1117             qq20             = _mm_mul_ps(iq2,jq0);
1118
1119             /* EWALD ELECTROSTATICS */
1120
1121             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1122             ewrt             = _mm_mul_ps(r20,ewtabscale);
1123             ewitab           = _mm_cvttps_epi32(ewrt);
1124             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1125             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1126                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1127                                          &ewtabF,&ewtabFn);
1128             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1129             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1130
1131             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1132
1133             fscal            = felec;
1134
1135             fscal            = _mm_and_ps(fscal,cutoff_mask);
1136
1137             /* Calculate temporary vectorial force */
1138             tx               = _mm_mul_ps(fscal,dx20);
1139             ty               = _mm_mul_ps(fscal,dy20);
1140             tz               = _mm_mul_ps(fscal,dz20);
1141
1142             /* Update vectorial force */
1143             fix2             = _mm_add_ps(fix2,tx);
1144             fiy2             = _mm_add_ps(fiy2,ty);
1145             fiz2             = _mm_add_ps(fiz2,tz);
1146
1147             fjx0             = _mm_add_ps(fjx0,tx);
1148             fjy0             = _mm_add_ps(fjy0,ty);
1149             fjz0             = _mm_add_ps(fjz0,tz);
1150
1151             }
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             if (gmx_mm_any_lt(rsq30,rcutoff2))
1158             {
1159
1160             r30              = _mm_mul_ps(rsq30,rinv30);
1161
1162             /* Compute parameters for interactions between i and j atoms */
1163             qq30             = _mm_mul_ps(iq3,jq0);
1164
1165             /* EWALD ELECTROSTATICS */
1166
1167             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1168             ewrt             = _mm_mul_ps(r30,ewtabscale);
1169             ewitab           = _mm_cvttps_epi32(ewrt);
1170             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1171             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1172                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1173                                          &ewtabF,&ewtabFn);
1174             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1175             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1176
1177             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1178
1179             fscal            = felec;
1180
1181             fscal            = _mm_and_ps(fscal,cutoff_mask);
1182
1183             /* Calculate temporary vectorial force */
1184             tx               = _mm_mul_ps(fscal,dx30);
1185             ty               = _mm_mul_ps(fscal,dy30);
1186             tz               = _mm_mul_ps(fscal,dz30);
1187
1188             /* Update vectorial force */
1189             fix3             = _mm_add_ps(fix3,tx);
1190             fiy3             = _mm_add_ps(fiy3,ty);
1191             fiz3             = _mm_add_ps(fiz3,tz);
1192
1193             fjx0             = _mm_add_ps(fjx0,tx);
1194             fjy0             = _mm_add_ps(fjy0,ty);
1195             fjz0             = _mm_add_ps(fjz0,tz);
1196
1197             }
1198
1199             fjptrA             = f+j_coord_offsetA;
1200             fjptrB             = f+j_coord_offsetB;
1201             fjptrC             = f+j_coord_offsetC;
1202             fjptrD             = f+j_coord_offsetD;
1203
1204             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1205
1206             /* Inner loop uses 147 flops */
1207         }
1208
1209         if(jidx<j_index_end)
1210         {
1211
1212             /* Get j neighbor index, and coordinate index */
1213             jnrlistA         = jjnr[jidx];
1214             jnrlistB         = jjnr[jidx+1];
1215             jnrlistC         = jjnr[jidx+2];
1216             jnrlistD         = jjnr[jidx+3];
1217             /* Sign of each element will be negative for non-real atoms.
1218              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1219              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1220              */
1221             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1222             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1223             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1224             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1225             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1226             j_coord_offsetA  = DIM*jnrA;
1227             j_coord_offsetB  = DIM*jnrB;
1228             j_coord_offsetC  = DIM*jnrC;
1229             j_coord_offsetD  = DIM*jnrD;
1230
1231             /* load j atom coordinates */
1232             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1233                                               x+j_coord_offsetC,x+j_coord_offsetD,
1234                                               &jx0,&jy0,&jz0);
1235
1236             /* Calculate displacement vector */
1237             dx00             = _mm_sub_ps(ix0,jx0);
1238             dy00             = _mm_sub_ps(iy0,jy0);
1239             dz00             = _mm_sub_ps(iz0,jz0);
1240             dx10             = _mm_sub_ps(ix1,jx0);
1241             dy10             = _mm_sub_ps(iy1,jy0);
1242             dz10             = _mm_sub_ps(iz1,jz0);
1243             dx20             = _mm_sub_ps(ix2,jx0);
1244             dy20             = _mm_sub_ps(iy2,jy0);
1245             dz20             = _mm_sub_ps(iz2,jz0);
1246             dx30             = _mm_sub_ps(ix3,jx0);
1247             dy30             = _mm_sub_ps(iy3,jy0);
1248             dz30             = _mm_sub_ps(iz3,jz0);
1249
1250             /* Calculate squared distance and things based on it */
1251             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1252             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1253             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1254             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1255
1256             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1257             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1258             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1259
1260             rinvsq00         = gmx_mm_inv_ps(rsq00);
1261             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1262             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1263             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1264
1265             /* Load parameters for j particles */
1266             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1267                                                               charge+jnrC+0,charge+jnrD+0);
1268             vdwjidx0A        = 2*vdwtype[jnrA+0];
1269             vdwjidx0B        = 2*vdwtype[jnrB+0];
1270             vdwjidx0C        = 2*vdwtype[jnrC+0];
1271             vdwjidx0D        = 2*vdwtype[jnrD+0];
1272
1273             fjx0             = _mm_setzero_ps();
1274             fjy0             = _mm_setzero_ps();
1275             fjz0             = _mm_setzero_ps();
1276
1277             /**************************
1278              * CALCULATE INTERACTIONS *
1279              **************************/
1280
1281             if (gmx_mm_any_lt(rsq00,rcutoff2))
1282             {
1283
1284             /* Compute parameters for interactions between i and j atoms */
1285             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1286                                          vdwparam+vdwioffset0+vdwjidx0B,
1287                                          vdwparam+vdwioffset0+vdwjidx0C,
1288                                          vdwparam+vdwioffset0+vdwjidx0D,
1289                                          &c6_00,&c12_00);
1290
1291             /* LENNARD-JONES DISPERSION/REPULSION */
1292
1293             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1294             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1295
1296             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1297
1298             fscal            = fvdw;
1299
1300             fscal            = _mm_and_ps(fscal,cutoff_mask);
1301
1302             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1303
1304             /* Calculate temporary vectorial force */
1305             tx               = _mm_mul_ps(fscal,dx00);
1306             ty               = _mm_mul_ps(fscal,dy00);
1307             tz               = _mm_mul_ps(fscal,dz00);
1308
1309             /* Update vectorial force */
1310             fix0             = _mm_add_ps(fix0,tx);
1311             fiy0             = _mm_add_ps(fiy0,ty);
1312             fiz0             = _mm_add_ps(fiz0,tz);
1313
1314             fjx0             = _mm_add_ps(fjx0,tx);
1315             fjy0             = _mm_add_ps(fjy0,ty);
1316             fjz0             = _mm_add_ps(fjz0,tz);
1317
1318             }
1319
1320             /**************************
1321              * CALCULATE INTERACTIONS *
1322              **************************/
1323
1324             if (gmx_mm_any_lt(rsq10,rcutoff2))
1325             {
1326
1327             r10              = _mm_mul_ps(rsq10,rinv10);
1328             r10              = _mm_andnot_ps(dummy_mask,r10);
1329
1330             /* Compute parameters for interactions between i and j atoms */
1331             qq10             = _mm_mul_ps(iq1,jq0);
1332
1333             /* EWALD ELECTROSTATICS */
1334
1335             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1336             ewrt             = _mm_mul_ps(r10,ewtabscale);
1337             ewitab           = _mm_cvttps_epi32(ewrt);
1338             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1339             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1340                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1341                                          &ewtabF,&ewtabFn);
1342             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1343             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1344
1345             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1346
1347             fscal            = felec;
1348
1349             fscal            = _mm_and_ps(fscal,cutoff_mask);
1350
1351             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1352
1353             /* Calculate temporary vectorial force */
1354             tx               = _mm_mul_ps(fscal,dx10);
1355             ty               = _mm_mul_ps(fscal,dy10);
1356             tz               = _mm_mul_ps(fscal,dz10);
1357
1358             /* Update vectorial force */
1359             fix1             = _mm_add_ps(fix1,tx);
1360             fiy1             = _mm_add_ps(fiy1,ty);
1361             fiz1             = _mm_add_ps(fiz1,tz);
1362
1363             fjx0             = _mm_add_ps(fjx0,tx);
1364             fjy0             = _mm_add_ps(fjy0,ty);
1365             fjz0             = _mm_add_ps(fjz0,tz);
1366
1367             }
1368
1369             /**************************
1370              * CALCULATE INTERACTIONS *
1371              **************************/
1372
1373             if (gmx_mm_any_lt(rsq20,rcutoff2))
1374             {
1375
1376             r20              = _mm_mul_ps(rsq20,rinv20);
1377             r20              = _mm_andnot_ps(dummy_mask,r20);
1378
1379             /* Compute parameters for interactions between i and j atoms */
1380             qq20             = _mm_mul_ps(iq2,jq0);
1381
1382             /* EWALD ELECTROSTATICS */
1383
1384             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1385             ewrt             = _mm_mul_ps(r20,ewtabscale);
1386             ewitab           = _mm_cvttps_epi32(ewrt);
1387             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1388             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1389                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1390                                          &ewtabF,&ewtabFn);
1391             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1392             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1393
1394             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1395
1396             fscal            = felec;
1397
1398             fscal            = _mm_and_ps(fscal,cutoff_mask);
1399
1400             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1401
1402             /* Calculate temporary vectorial force */
1403             tx               = _mm_mul_ps(fscal,dx20);
1404             ty               = _mm_mul_ps(fscal,dy20);
1405             tz               = _mm_mul_ps(fscal,dz20);
1406
1407             /* Update vectorial force */
1408             fix2             = _mm_add_ps(fix2,tx);
1409             fiy2             = _mm_add_ps(fiy2,ty);
1410             fiz2             = _mm_add_ps(fiz2,tz);
1411
1412             fjx0             = _mm_add_ps(fjx0,tx);
1413             fjy0             = _mm_add_ps(fjy0,ty);
1414             fjz0             = _mm_add_ps(fjz0,tz);
1415
1416             }
1417
1418             /**************************
1419              * CALCULATE INTERACTIONS *
1420              **************************/
1421
1422             if (gmx_mm_any_lt(rsq30,rcutoff2))
1423             {
1424
1425             r30              = _mm_mul_ps(rsq30,rinv30);
1426             r30              = _mm_andnot_ps(dummy_mask,r30);
1427
1428             /* Compute parameters for interactions between i and j atoms */
1429             qq30             = _mm_mul_ps(iq3,jq0);
1430
1431             /* EWALD ELECTROSTATICS */
1432
1433             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1434             ewrt             = _mm_mul_ps(r30,ewtabscale);
1435             ewitab           = _mm_cvttps_epi32(ewrt);
1436             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1437             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1438                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1439                                          &ewtabF,&ewtabFn);
1440             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1441             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1442
1443             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1444
1445             fscal            = felec;
1446
1447             fscal            = _mm_and_ps(fscal,cutoff_mask);
1448
1449             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1450
1451             /* Calculate temporary vectorial force */
1452             tx               = _mm_mul_ps(fscal,dx30);
1453             ty               = _mm_mul_ps(fscal,dy30);
1454             tz               = _mm_mul_ps(fscal,dz30);
1455
1456             /* Update vectorial force */
1457             fix3             = _mm_add_ps(fix3,tx);
1458             fiy3             = _mm_add_ps(fiy3,ty);
1459             fiz3             = _mm_add_ps(fiz3,tz);
1460
1461             fjx0             = _mm_add_ps(fjx0,tx);
1462             fjy0             = _mm_add_ps(fjy0,ty);
1463             fjz0             = _mm_add_ps(fjz0,tz);
1464
1465             }
1466
1467             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1468             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1469             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1470             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1471
1472             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1473
1474             /* Inner loop uses 150 flops */
1475         }
1476
1477         /* End of innermost loop */
1478
1479         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1480                                               f+i_coord_offset,fshift+i_shift_offset);
1481
1482         /* Increment number of inner iterations */
1483         inneriter                  += j_index_end - j_index_start;
1484
1485         /* Outer loop uses 24 flops */
1486     }
1487
1488     /* Increment number of outer iterations */
1489     outeriter        += nri;
1490
1491     /* Update outer/inner flops */
1492
1493     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1494 }