src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSw_VdwLJSw_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_ElecEwSw_VdwLJSw_GeomW4P1_VF_sse4_1_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwLJSw_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     __m128i          ewitab;
 106     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     real             *ewtab;
 108     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 109     real             rswitch_scalar,d_scalar;
 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     rswitch_scalar   = fr->rcoulomb_switch;
 149     rswitch          = _mm_set1_ps(rswitch_scalar);
 150     /* Setup switch parameters */
 151     d_scalar         = rcutoff_scalar-rswitch_scalar;
 152     d                = _mm_set1_ps(d_scalar);
 153     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 154     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 155     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 156     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 157     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 158     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 159
 160     /* Avoid stupid compiler warnings */
 161     jnrA = jnrB = jnrC = jnrD = 0;
 162     j_coord_offsetA = 0;
 163     j_coord_offsetB = 0;
 164     j_coord_offsetC = 0;
 165     j_coord_offsetD = 0;
 166
 167     outeriter        = 0;
 168     inneriter        = 0;
 169
 170     for(iidx=0;iidx<4*DIM;iidx++)
 171     {
 172         scratch[iidx] = 0.0;
 173     }
 174
 175     /* Start outer loop over neighborlists */
 176     for(iidx=0; iidx<nri; iidx++)
 177     {
 178         /* Load shift vector for this list */
 179         i_shift_offset   = DIM*shiftidx[iidx];
 180
 181         /* Load limits for loop over neighbors */
 182         j_index_start    = jindex[iidx];
 183         j_index_end      = jindex[iidx+1];
 184
 185         /* Get outer coordinate index */
 186         inr              = iinr[iidx];
 187         i_coord_offset   = DIM*inr;
 188
 189         /* Load i particle coords and add shift vector */
 190         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 191                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 192
 193         fix0             = _mm_setzero_ps();
 194         fiy0             = _mm_setzero_ps();
 195         fiz0             = _mm_setzero_ps();
 196         fix1             = _mm_setzero_ps();
 197         fiy1             = _mm_setzero_ps();
 198         fiz1             = _mm_setzero_ps();
 199         fix2             = _mm_setzero_ps();
 200         fiy2             = _mm_setzero_ps();
 201         fiz2             = _mm_setzero_ps();
 202         fix3             = _mm_setzero_ps();
 203         fiy3             = _mm_setzero_ps();
 204         fiz3             = _mm_setzero_ps();
 205
 206         /* Reset potential sums */
 207         velecsum         = _mm_setzero_ps();
 208         vvdwsum          = _mm_setzero_ps();
 209
 210         /* Start inner kernel loop */
 211         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 212         {
 213
 214             /* Get j neighbor index, and coordinate index */
 215             jnrA             = jjnr[jidx];
 216             jnrB             = jjnr[jidx+1];
 217             jnrC             = jjnr[jidx+2];
 218             jnrD             = jjnr[jidx+3];
 219             j_coord_offsetA  = DIM*jnrA;
 220             j_coord_offsetB  = DIM*jnrB;
 221             j_coord_offsetC  = DIM*jnrC;
 222             j_coord_offsetD  = DIM*jnrD;
 223
 224             /* load j atom coordinates */
 225             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 226                                               x+j_coord_offsetC,x+j_coord_offsetD,
 227                                               &jx0,&jy0,&jz0);
 228
 229             /* Calculate displacement vector */
 230             dx00             = _mm_sub_ps(ix0,jx0);
 231             dy00             = _mm_sub_ps(iy0,jy0);
 232             dz00             = _mm_sub_ps(iz0,jz0);
 233             dx10             = _mm_sub_ps(ix1,jx0);
 234             dy10             = _mm_sub_ps(iy1,jy0);
 235             dz10             = _mm_sub_ps(iz1,jz0);
 236             dx20             = _mm_sub_ps(ix2,jx0);
 237             dy20             = _mm_sub_ps(iy2,jy0);
 238             dz20             = _mm_sub_ps(iz2,jz0);
 239             dx30             = _mm_sub_ps(ix3,jx0);
 240             dy30             = _mm_sub_ps(iy3,jy0);
 241             dz30             = _mm_sub_ps(iz3,jz0);
 242
 243             /* Calculate squared distance and things based on it */
 244             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 245             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 246             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 247             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 248
 249             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 250             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 251             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 252             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 253
 254             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 255             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 256             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 257             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 258
 259             /* Load parameters for j particles */
 260             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 261                                                               charge+jnrC+0,charge+jnrD+0);
 262             vdwjidx0A        = 2*vdwtype[jnrA+0];
 263             vdwjidx0B        = 2*vdwtype[jnrB+0];
 264             vdwjidx0C        = 2*vdwtype[jnrC+0];
 265             vdwjidx0D        = 2*vdwtype[jnrD+0];
 266
 267             fjx0             = _mm_setzero_ps();
 268             fjy0             = _mm_setzero_ps();
 269             fjz0             = _mm_setzero_ps();
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             if (gmx_mm_any_lt(rsq00,rcutoff2))
 276             {
 277
 278             r00              = _mm_mul_ps(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 282                                          vdwparam+vdwioffset0+vdwjidx0B,
 283                                          vdwparam+vdwioffset0+vdwjidx0C,
 284                                          vdwparam+vdwioffset0+vdwjidx0D,
 285                                          &c6_00,&c12_00);
 286
 287             /* LENNARD-JONES DISPERSION/REPULSION */
 288
 289             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 290             vvdw6            = _mm_mul_ps(c6_00,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             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 294
 295             d                = _mm_sub_ps(r00,rswitch);
 296             d                = _mm_max_ps(d,_mm_setzero_ps());
 297             d2               = _mm_mul_ps(d,d);
 298             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 299
 300             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 301
 302             /* Evaluate switch function */
 303             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 304             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 305             vvdw             = _mm_mul_ps(vvdw,sw);
 306             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 310             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 311
 312             fscal            = fvdw;
 313
 314             fscal            = _mm_and_ps(fscal,cutoff_mask);
 315
 316             /* Calculate temporary vectorial force */
 317             tx               = _mm_mul_ps(fscal,dx00);
 318             ty               = _mm_mul_ps(fscal,dy00);
 319             tz               = _mm_mul_ps(fscal,dz00);
 320
 321             /* Update vectorial force */
 322             fix0             = _mm_add_ps(fix0,tx);
 323             fiy0             = _mm_add_ps(fiy0,ty);
 324             fiz0             = _mm_add_ps(fiz0,tz);
 325
 326             fjx0             = _mm_add_ps(fjx0,tx);
 327             fjy0             = _mm_add_ps(fjy0,ty);
 328             fjz0             = _mm_add_ps(fjz0,tz);
 329
 330             }
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             if (gmx_mm_any_lt(rsq10,rcutoff2))
 337             {
 338
 339             r10              = _mm_mul_ps(rsq10,rinv10);
 340
 341             /* Compute parameters for interactions between i and j atoms */
 342             qq10             = _mm_mul_ps(iq1,jq0);
 343
 344             /* EWALD ELECTROSTATICS */
 345
 346             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 347             ewrt             = _mm_mul_ps(r10,ewtabscale);
 348             ewitab           = _mm_cvttps_epi32(ewrt);
 349             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 350             ewitab           = _mm_slli_epi32(ewitab,2);
 351             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 352             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 353             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 354             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 355             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 356             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 357             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 358             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 359             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 360
 361             d                = _mm_sub_ps(r10,rswitch);
 362             d                = _mm_max_ps(d,_mm_setzero_ps());
 363             d2               = _mm_mul_ps(d,d);
 364             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 365
 366             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 367
 368             /* Evaluate switch function */
 369             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 370             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 371             velec            = _mm_mul_ps(velec,sw);
 372             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velec            = _mm_and_ps(velec,cutoff_mask);
 376             velecsum         = _mm_add_ps(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             fscal            = _mm_and_ps(fscal,cutoff_mask);
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm_mul_ps(fscal,dx10);
 384             ty               = _mm_mul_ps(fscal,dy10);
 385             tz               = _mm_mul_ps(fscal,dz10);
 386
 387             /* Update vectorial force */
 388             fix1             = _mm_add_ps(fix1,tx);
 389             fiy1             = _mm_add_ps(fiy1,ty);
 390             fiz1             = _mm_add_ps(fiz1,tz);
 391
 392             fjx0             = _mm_add_ps(fjx0,tx);
 393             fjy0             = _mm_add_ps(fjy0,ty);
 394             fjz0             = _mm_add_ps(fjz0,tz);
 395
 396             }
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             if (gmx_mm_any_lt(rsq20,rcutoff2))
 403             {
 404
 405             r20              = _mm_mul_ps(rsq20,rinv20);
 406
 407             /* Compute parameters for interactions between i and j atoms */
 408             qq20             = _mm_mul_ps(iq2,jq0);
 409
 410             /* EWALD ELECTROSTATICS */
 411
 412             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 413             ewrt             = _mm_mul_ps(r20,ewtabscale);
 414             ewitab           = _mm_cvttps_epi32(ewrt);
 415             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 416             ewitab           = _mm_slli_epi32(ewitab,2);
 417             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 418             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 419             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 420             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 421             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 422             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 423             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 424             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 425             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 426
 427             d                = _mm_sub_ps(r20,rswitch);
 428             d                = _mm_max_ps(d,_mm_setzero_ps());
 429             d2               = _mm_mul_ps(d,d);
 430             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 431
 432             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 433
 434             /* Evaluate switch function */
 435             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 436             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 437             velec            = _mm_mul_ps(velec,sw);
 438             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 439
 440             /* Update potential sum for this i atom from the interaction with this j atom. */
 441             velec            = _mm_and_ps(velec,cutoff_mask);
 442             velecsum         = _mm_add_ps(velecsum,velec);
 443
 444             fscal            = felec;
 445
 446             fscal            = _mm_and_ps(fscal,cutoff_mask);
 447
 448             /* Calculate temporary vectorial force */
 449             tx               = _mm_mul_ps(fscal,dx20);
 450             ty               = _mm_mul_ps(fscal,dy20);
 451             tz               = _mm_mul_ps(fscal,dz20);
 452
 453             /* Update vectorial force */
 454             fix2             = _mm_add_ps(fix2,tx);
 455             fiy2             = _mm_add_ps(fiy2,ty);
 456             fiz2             = _mm_add_ps(fiz2,tz);
 457
 458             fjx0             = _mm_add_ps(fjx0,tx);
 459             fjy0             = _mm_add_ps(fjy0,ty);
 460             fjz0             = _mm_add_ps(fjz0,tz);
 461
 462             }
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             if (gmx_mm_any_lt(rsq30,rcutoff2))
 469             {
 470
 471             r30              = _mm_mul_ps(rsq30,rinv30);
 472
 473             /* Compute parameters for interactions between i and j atoms */
 474             qq30             = _mm_mul_ps(iq3,jq0);
 475
 476             /* EWALD ELECTROSTATICS */
 477
 478             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 479             ewrt             = _mm_mul_ps(r30,ewtabscale);
 480             ewitab           = _mm_cvttps_epi32(ewrt);
 481             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 482             ewitab           = _mm_slli_epi32(ewitab,2);
 483             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 484             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 485             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 486             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 487             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 488             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 489             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 490             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 491             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 492
 493             d                = _mm_sub_ps(r30,rswitch);
 494             d                = _mm_max_ps(d,_mm_setzero_ps());
 495             d2               = _mm_mul_ps(d,d);
 496             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 497
 498             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 499
 500             /* Evaluate switch function */
 501             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 502             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
 503             velec            = _mm_mul_ps(velec,sw);
 504             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 505
 506             /* Update potential sum for this i atom from the interaction with this j atom. */
 507             velec            = _mm_and_ps(velec,cutoff_mask);
 508             velecsum         = _mm_add_ps(velecsum,velec);
 509
 510             fscal            = felec;
 511
 512             fscal            = _mm_and_ps(fscal,cutoff_mask);
 513
 514             /* Calculate temporary vectorial force */
 515             tx               = _mm_mul_ps(fscal,dx30);
 516             ty               = _mm_mul_ps(fscal,dy30);
 517             tz               = _mm_mul_ps(fscal,dz30);
 518
 519             /* Update vectorial force */
 520             fix3             = _mm_add_ps(fix3,tx);
 521             fiy3             = _mm_add_ps(fiy3,ty);
 522             fiz3             = _mm_add_ps(fiz3,tz);
 523
 524             fjx0             = _mm_add_ps(fjx0,tx);
 525             fjy0             = _mm_add_ps(fjy0,ty);
 526             fjz0             = _mm_add_ps(fjz0,tz);
 527
 528             }
 529
 530             fjptrA             = f+j_coord_offsetA;
 531             fjptrB             = f+j_coord_offsetB;
 532             fjptrC             = f+j_coord_offsetC;
 533             fjptrD             = f+j_coord_offsetD;
 534
 535             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 536
 537             /* Inner loop uses 254 flops */
 538         }
 539
 540         if(jidx<j_index_end)
 541         {
 542
 543             /* Get j neighbor index, and coordinate index */
 544             jnrlistA         = jjnr[jidx];
 545             jnrlistB         = jjnr[jidx+1];
 546             jnrlistC         = jjnr[jidx+2];
 547             jnrlistD         = jjnr[jidx+3];
 548             /* Sign of each element will be negative for non-real atoms.
 549              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 550              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 551              */
 552             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 553             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 554             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 555             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 556             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 557             j_coord_offsetA  = DIM*jnrA;
 558             j_coord_offsetB  = DIM*jnrB;
 559             j_coord_offsetC  = DIM*jnrC;
 560             j_coord_offsetD  = DIM*jnrD;
 561
 562             /* load j atom coordinates */
 563             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 564                                               x+j_coord_offsetC,x+j_coord_offsetD,
 565                                               &jx0,&jy0,&jz0);
 566
 567             /* Calculate displacement vector */
 568             dx00             = _mm_sub_ps(ix0,jx0);
 569             dy00             = _mm_sub_ps(iy0,jy0);
 570             dz00             = _mm_sub_ps(iz0,jz0);
 571             dx10             = _mm_sub_ps(ix1,jx0);
 572             dy10             = _mm_sub_ps(iy1,jy0);
 573             dz10             = _mm_sub_ps(iz1,jz0);
 574             dx20             = _mm_sub_ps(ix2,jx0);
 575             dy20             = _mm_sub_ps(iy2,jy0);
 576             dz20             = _mm_sub_ps(iz2,jz0);
 577             dx30             = _mm_sub_ps(ix3,jx0);
 578             dy30             = _mm_sub_ps(iy3,jy0);
 579             dz30             = _mm_sub_ps(iz3,jz0);
 580
 581             /* Calculate squared distance and things based on it */
 582             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 583             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 584             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 585             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 586
 587             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 588             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 589             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 590             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 591
 592             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 593             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 594             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 595             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 596
 597             /* Load parameters for j particles */
 598             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 599                                                               charge+jnrC+0,charge+jnrD+0);
 600             vdwjidx0A        = 2*vdwtype[jnrA+0];
 601             vdwjidx0B        = 2*vdwtype[jnrB+0];
 602             vdwjidx0C        = 2*vdwtype[jnrC+0];
 603             vdwjidx0D        = 2*vdwtype[jnrD+0];
 604
 605             fjx0             = _mm_setzero_ps();
 606             fjy0             = _mm_setzero_ps();
 607             fjz0             = _mm_setzero_ps();
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             if (gmx_mm_any_lt(rsq00,rcutoff2))
 614             {
 615
 616             r00              = _mm_mul_ps(rsq00,rinv00);
 617             r00              = _mm_andnot_ps(dummy_mask,r00);
 618
 619             /* Compute parameters for interactions between i and j atoms */
 620             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 621                                          vdwparam+vdwioffset0+vdwjidx0B,
 622                                          vdwparam+vdwioffset0+vdwjidx0C,
 623                                          vdwparam+vdwioffset0+vdwjidx0D,
 624                                          &c6_00,&c12_00);
 625
 626             /* LENNARD-JONES DISPERSION/REPULSION */
 627
 628             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 629             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 630             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 631             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 632             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 633
 634             d                = _mm_sub_ps(r00,rswitch);
 635             d                = _mm_max_ps(d,_mm_setzero_ps());
 636             d2               = _mm_mul_ps(d,d);
 637             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 638
 639             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 640
 641             /* Evaluate switch function */
 642             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 643             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 644             vvdw             = _mm_mul_ps(vvdw,sw);
 645             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 646
 647             /* Update potential sum for this i atom from the interaction with this j atom. */
 648             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 649             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 650             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 651
 652             fscal            = fvdw;
 653
 654             fscal            = _mm_and_ps(fscal,cutoff_mask);
 655
 656             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 657
 658             /* Calculate temporary vectorial force */
 659             tx               = _mm_mul_ps(fscal,dx00);
 660             ty               = _mm_mul_ps(fscal,dy00);
 661             tz               = _mm_mul_ps(fscal,dz00);
 662
 663             /* Update vectorial force */
 664             fix0             = _mm_add_ps(fix0,tx);
 665             fiy0             = _mm_add_ps(fiy0,ty);
 666             fiz0             = _mm_add_ps(fiz0,tz);
 667
 668             fjx0             = _mm_add_ps(fjx0,tx);
 669             fjy0             = _mm_add_ps(fjy0,ty);
 670             fjz0             = _mm_add_ps(fjz0,tz);
 671
 672             }
 673
 674             /**************************
 675              * CALCULATE INTERACTIONS *
 676              **************************/
 677
 678             if (gmx_mm_any_lt(rsq10,rcutoff2))
 679             {
 680
 681             r10              = _mm_mul_ps(rsq10,rinv10);
 682             r10              = _mm_andnot_ps(dummy_mask,r10);
 683
 684             /* Compute parameters for interactions between i and j atoms */
 685             qq10             = _mm_mul_ps(iq1,jq0);
 686
 687             /* EWALD ELECTROSTATICS */
 688
 689             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 690             ewrt             = _mm_mul_ps(r10,ewtabscale);
 691             ewitab           = _mm_cvttps_epi32(ewrt);
 692             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 693             ewitab           = _mm_slli_epi32(ewitab,2);
 694             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 695             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 696             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 697             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 698             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 699             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 700             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 701             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 702             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 703
 704             d                = _mm_sub_ps(r10,rswitch);
 705             d                = _mm_max_ps(d,_mm_setzero_ps());
 706             d2               = _mm_mul_ps(d,d);
 707             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 708
 709             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 710
 711             /* Evaluate switch function */
 712             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 713             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 714             velec            = _mm_mul_ps(velec,sw);
 715             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 716
 717             /* Update potential sum for this i atom from the interaction with this j atom. */
 718             velec            = _mm_and_ps(velec,cutoff_mask);
 719             velec            = _mm_andnot_ps(dummy_mask,velec);
 720             velecsum         = _mm_add_ps(velecsum,velec);
 721
 722             fscal            = felec;
 723
 724             fscal            = _mm_and_ps(fscal,cutoff_mask);
 725
 726             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 727
 728             /* Calculate temporary vectorial force */
 729             tx               = _mm_mul_ps(fscal,dx10);
 730             ty               = _mm_mul_ps(fscal,dy10);
 731             tz               = _mm_mul_ps(fscal,dz10);
 732
 733             /* Update vectorial force */
 734             fix1             = _mm_add_ps(fix1,tx);
 735             fiy1             = _mm_add_ps(fiy1,ty);
 736             fiz1             = _mm_add_ps(fiz1,tz);
 737
 738             fjx0             = _mm_add_ps(fjx0,tx);
 739             fjy0             = _mm_add_ps(fjy0,ty);
 740             fjz0             = _mm_add_ps(fjz0,tz);
 741
 742             }
 743
 744             /**************************
 745              * CALCULATE INTERACTIONS *
 746              **************************/
 747
 748             if (gmx_mm_any_lt(rsq20,rcutoff2))
 749             {
 750
 751             r20              = _mm_mul_ps(rsq20,rinv20);
 752             r20              = _mm_andnot_ps(dummy_mask,r20);
 753
 754             /* Compute parameters for interactions between i and j atoms */
 755             qq20             = _mm_mul_ps(iq2,jq0);
 756
 757             /* EWALD ELECTROSTATICS */
 758
 759             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 760             ewrt             = _mm_mul_ps(r20,ewtabscale);
 761             ewitab           = _mm_cvttps_epi32(ewrt);
 762             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 763             ewitab           = _mm_slli_epi32(ewitab,2);
 764             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 765             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 766             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 767             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 768             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 769             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 770             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 771             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 772             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 773
 774             d                = _mm_sub_ps(r20,rswitch);
 775             d                = _mm_max_ps(d,_mm_setzero_ps());
 776             d2               = _mm_mul_ps(d,d);
 777             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 778
 779             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 780
 781             /* Evaluate switch function */
 782             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 783             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 784             velec            = _mm_mul_ps(velec,sw);
 785             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 786
 787             /* Update potential sum for this i atom from the interaction with this j atom. */
 788             velec            = _mm_and_ps(velec,cutoff_mask);
 789             velec            = _mm_andnot_ps(dummy_mask,velec);
 790             velecsum         = _mm_add_ps(velecsum,velec);
 791
 792             fscal            = felec;
 793
 794             fscal            = _mm_and_ps(fscal,cutoff_mask);
 795
 796             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 797
 798             /* Calculate temporary vectorial force */
 799             tx               = _mm_mul_ps(fscal,dx20);
 800             ty               = _mm_mul_ps(fscal,dy20);
 801             tz               = _mm_mul_ps(fscal,dz20);
 802
 803             /* Update vectorial force */
 804             fix2             = _mm_add_ps(fix2,tx);
 805             fiy2             = _mm_add_ps(fiy2,ty);
 806             fiz2             = _mm_add_ps(fiz2,tz);
 807
 808             fjx0             = _mm_add_ps(fjx0,tx);
 809             fjy0             = _mm_add_ps(fjy0,ty);
 810             fjz0             = _mm_add_ps(fjz0,tz);
 811
 812             }
 813
 814             /**************************
 815              * CALCULATE INTERACTIONS *
 816              **************************/
 817
 818             if (gmx_mm_any_lt(rsq30,rcutoff2))
 819             {
 820
 821             r30              = _mm_mul_ps(rsq30,rinv30);
 822             r30              = _mm_andnot_ps(dummy_mask,r30);
 823
 824             /* Compute parameters for interactions between i and j atoms */
 825             qq30             = _mm_mul_ps(iq3,jq0);
 826
 827             /* EWALD ELECTROSTATICS */
 828
 829             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 830             ewrt             = _mm_mul_ps(r30,ewtabscale);
 831             ewitab           = _mm_cvttps_epi32(ewrt);
 832             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 833             ewitab           = _mm_slli_epi32(ewitab,2);
 834             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 835             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 836             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 837             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 838             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 839             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 840             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 841             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 842             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 843
 844             d                = _mm_sub_ps(r30,rswitch);
 845             d                = _mm_max_ps(d,_mm_setzero_ps());
 846             d2               = _mm_mul_ps(d,d);
 847             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 848
 849             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 850
 851             /* Evaluate switch function */
 852             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 853             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
 854             velec            = _mm_mul_ps(velec,sw);
 855             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 856
 857             /* Update potential sum for this i atom from the interaction with this j atom. */
 858             velec            = _mm_and_ps(velec,cutoff_mask);
 859             velec            = _mm_andnot_ps(dummy_mask,velec);
 860             velecsum         = _mm_add_ps(velecsum,velec);
 861
 862             fscal            = felec;
 863
 864             fscal            = _mm_and_ps(fscal,cutoff_mask);
 865
 866             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 867
 868             /* Calculate temporary vectorial force */
 869             tx               = _mm_mul_ps(fscal,dx30);
 870             ty               = _mm_mul_ps(fscal,dy30);
 871             tz               = _mm_mul_ps(fscal,dz30);
 872
 873             /* Update vectorial force */
 874             fix3             = _mm_add_ps(fix3,tx);
 875             fiy3             = _mm_add_ps(fiy3,ty);
 876             fiz3             = _mm_add_ps(fiz3,tz);
 877
 878             fjx0             = _mm_add_ps(fjx0,tx);
 879             fjy0             = _mm_add_ps(fjy0,ty);
 880             fjz0             = _mm_add_ps(fjz0,tz);
 881
 882             }
 883
 884             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 885             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 886             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 887             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 888
 889             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 890
 891             /* Inner loop uses 258 flops */
 892         }
 893
 894         /* End of innermost loop */
 895
 896         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 897                                               f+i_coord_offset,fshift+i_shift_offset);
 898
 899         ggid                        = gid[iidx];
 900         /* Update potential energies */
 901         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 902         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 903
 904         /* Increment number of inner iterations */
 905         inneriter                  += j_index_end - j_index_start;
 906
 907         /* Outer loop uses 26 flops */
 908     }
 909
 910     /* Increment number of outer iterations */
 911     outeriter        += nri;
 912
 913     /* Update outer/inner flops */
 914
 915     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*258);
 916 }
 917 /*
 918  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_sse4_1_single
 919  * Electrostatics interaction: Ewald
 920  * VdW interaction:            LennardJones
 921  * Geometry:                   Water4-Particle
 922  * Calculate force/pot:        Force
 923  */
 924 void
 925 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_sse4_1_single
 926                     (t_nblist                    * gmx_restrict       nlist,
 927                      rvec                        * gmx_restrict          xx,
 928                      rvec                        * gmx_restrict          ff,
 929                      t_forcerec                  * gmx_restrict          fr,
 930                      t_mdatoms                   * gmx_restrict     mdatoms,
 931                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 932                      t_nrnb                      * gmx_restrict        nrnb)
 933 {
 934     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 935      * just 0 for non-waters.
 936      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 937      * jnr indices corresponding to data put in the four positions in the SIMD register.
 938      */
 939     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 940     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 941     int              jnrA,jnrB,jnrC,jnrD;
 942     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 943     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 944     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 945     real             rcutoff_scalar;
 946     real             *shiftvec,*fshift,*x,*f;
 947     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 948     real             scratch[4*DIM];
 949     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 950     int              vdwioffset0;
 951     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 952     int              vdwioffset1;
 953     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 954     int              vdwioffset2;
 955     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 956     int              vdwioffset3;
 957     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 958     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 959     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 960     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 961     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 962     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 963     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 964     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 965     real             *charge;
 966     int              nvdwtype;
 967     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 968     int              *vdwtype;
 969     real             *vdwparam;
 970     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 971     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 972     __m128i          ewitab;
 973     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 974     real             *ewtab;
 975     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 976     real             rswitch_scalar,d_scalar;
 977     __m128           dummy_mask,cutoff_mask;
 978     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 979     __m128           one     = _mm_set1_ps(1.0);
 980     __m128           two     = _mm_set1_ps(2.0);
 981     x                = xx[0];
 982     f                = ff[0];
 983
 984     nri              = nlist->nri;
 985     iinr             = nlist->iinr;
 986     jindex           = nlist->jindex;
 987     jjnr             = nlist->jjnr;
 988     shiftidx         = nlist->shift;
 989     gid              = nlist->gid;
 990     shiftvec         = fr->shift_vec[0];
 991     fshift           = fr->fshift[0];
 992     facel            = _mm_set1_ps(fr->epsfac);
 993     charge           = mdatoms->chargeA;
 994     nvdwtype         = fr->ntype;
 995     vdwparam         = fr->nbfp;
 996     vdwtype          = mdatoms->typeA;
 997
 998     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 999     ewtab            = fr->ic->tabq_coul_FDV0;
1000     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1001     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1002
1003     /* Setup water-specific parameters */
1004     inr              = nlist->iinr[0];
1005     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1006     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1007     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1008     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1009
1010     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1011     rcutoff_scalar   = fr->rcoulomb;
1012     rcutoff          = _mm_set1_ps(rcutoff_scalar);
1013     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
1014
1015     rswitch_scalar   = fr->rcoulomb_switch;
1016     rswitch          = _mm_set1_ps(rswitch_scalar);
1017     /* Setup switch parameters */
1018     d_scalar         = rcutoff_scalar-rswitch_scalar;
1019     d                = _mm_set1_ps(d_scalar);
1020     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
1021     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1022     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1023     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
1024     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1025     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1026
1027     /* Avoid stupid compiler warnings */
1028     jnrA = jnrB = jnrC = jnrD = 0;
1029     j_coord_offsetA = 0;
1030     j_coord_offsetB = 0;
1031     j_coord_offsetC = 0;
1032     j_coord_offsetD = 0;
1033
1034     outeriter        = 0;
1035     inneriter        = 0;
1036
1037     for(iidx=0;iidx<4*DIM;iidx++)
1038     {
1039         scratch[iidx] = 0.0;
1040     }
1041
1042     /* Start outer loop over neighborlists */
1043     for(iidx=0; iidx<nri; iidx++)
1044     {
1045         /* Load shift vector for this list */
1046         i_shift_offset   = DIM*shiftidx[iidx];
1047
1048         /* Load limits for loop over neighbors */
1049         j_index_start    = jindex[iidx];
1050         j_index_end      = jindex[iidx+1];
1051
1052         /* Get outer coordinate index */
1053         inr              = iinr[iidx];
1054         i_coord_offset   = DIM*inr;
1055
1056         /* Load i particle coords and add shift vector */
1057         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1058                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1059
1060         fix0             = _mm_setzero_ps();
1061         fiy0             = _mm_setzero_ps();
1062         fiz0             = _mm_setzero_ps();
1063         fix1             = _mm_setzero_ps();
1064         fiy1             = _mm_setzero_ps();
1065         fiz1             = _mm_setzero_ps();
1066         fix2             = _mm_setzero_ps();
1067         fiy2             = _mm_setzero_ps();
1068         fiz2             = _mm_setzero_ps();
1069         fix3             = _mm_setzero_ps();
1070         fiy3             = _mm_setzero_ps();
1071         fiz3             = _mm_setzero_ps();
1072
1073         /* Start inner kernel loop */
1074         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1075         {
1076
1077             /* Get j neighbor index, and coordinate index */
1078             jnrA             = jjnr[jidx];
1079             jnrB             = jjnr[jidx+1];
1080             jnrC             = jjnr[jidx+2];
1081             jnrD             = jjnr[jidx+3];
1082             j_coord_offsetA  = DIM*jnrA;
1083             j_coord_offsetB  = DIM*jnrB;
1084             j_coord_offsetC  = DIM*jnrC;
1085             j_coord_offsetD  = DIM*jnrD;
1086
1087             /* load j atom coordinates */
1088             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1089                                               x+j_coord_offsetC,x+j_coord_offsetD,
1090                                               &jx0,&jy0,&jz0);
1091
1092             /* Calculate displacement vector */
1093             dx00             = _mm_sub_ps(ix0,jx0);
1094             dy00             = _mm_sub_ps(iy0,jy0);
1095             dz00             = _mm_sub_ps(iz0,jz0);
1096             dx10             = _mm_sub_ps(ix1,jx0);
1097             dy10             = _mm_sub_ps(iy1,jy0);
1098             dz10             = _mm_sub_ps(iz1,jz0);
1099             dx20             = _mm_sub_ps(ix2,jx0);
1100             dy20             = _mm_sub_ps(iy2,jy0);
1101             dz20             = _mm_sub_ps(iz2,jz0);
1102             dx30             = _mm_sub_ps(ix3,jx0);
1103             dy30             = _mm_sub_ps(iy3,jy0);
1104             dz30             = _mm_sub_ps(iz3,jz0);
1105
1106             /* Calculate squared distance and things based on it */
1107             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1108             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1109             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1110             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1111
1112             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1113             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1114             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1115             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1116
1117             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1118             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1119             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1120             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1121
1122             /* Load parameters for j particles */
1123             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1124                                                               charge+jnrC+0,charge+jnrD+0);
1125             vdwjidx0A        = 2*vdwtype[jnrA+0];
1126             vdwjidx0B        = 2*vdwtype[jnrB+0];
1127             vdwjidx0C        = 2*vdwtype[jnrC+0];
1128             vdwjidx0D        = 2*vdwtype[jnrD+0];
1129
1130             fjx0             = _mm_setzero_ps();
1131             fjy0             = _mm_setzero_ps();
1132             fjz0             = _mm_setzero_ps();
1133
1134             /**************************
1135              * CALCULATE INTERACTIONS *
1136              **************************/
1137
1138             if (gmx_mm_any_lt(rsq00,rcutoff2))
1139             {
1140
1141             r00              = _mm_mul_ps(rsq00,rinv00);
1142
1143             /* Compute parameters for interactions between i and j atoms */
1144             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1145                                          vdwparam+vdwioffset0+vdwjidx0B,
1146                                          vdwparam+vdwioffset0+vdwjidx0C,
1147                                          vdwparam+vdwioffset0+vdwjidx0D,
1148                                          &c6_00,&c12_00);
1149
1150             /* LENNARD-JONES DISPERSION/REPULSION */
1151
1152             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1153             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1154             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1155             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1156             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1157
1158             d                = _mm_sub_ps(r00,rswitch);
1159             d                = _mm_max_ps(d,_mm_setzero_ps());
1160             d2               = _mm_mul_ps(d,d);
1161             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1162
1163             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1164
1165             /* Evaluate switch function */
1166             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1167             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1168             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1169
1170             fscal            = fvdw;
1171
1172             fscal            = _mm_and_ps(fscal,cutoff_mask);
1173
1174             /* Calculate temporary vectorial force */
1175             tx               = _mm_mul_ps(fscal,dx00);
1176             ty               = _mm_mul_ps(fscal,dy00);
1177             tz               = _mm_mul_ps(fscal,dz00);
1178
1179             /* Update vectorial force */
1180             fix0             = _mm_add_ps(fix0,tx);
1181             fiy0             = _mm_add_ps(fiy0,ty);
1182             fiz0             = _mm_add_ps(fiz0,tz);
1183
1184             fjx0             = _mm_add_ps(fjx0,tx);
1185             fjy0             = _mm_add_ps(fjy0,ty);
1186             fjz0             = _mm_add_ps(fjz0,tz);
1187
1188             }
1189
1190             /**************************
1191              * CALCULATE INTERACTIONS *
1192              **************************/
1193
1194             if (gmx_mm_any_lt(rsq10,rcutoff2))
1195             {
1196
1197             r10              = _mm_mul_ps(rsq10,rinv10);
1198
1199             /* Compute parameters for interactions between i and j atoms */
1200             qq10             = _mm_mul_ps(iq1,jq0);
1201
1202             /* EWALD ELECTROSTATICS */
1203
1204             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1205             ewrt             = _mm_mul_ps(r10,ewtabscale);
1206             ewitab           = _mm_cvttps_epi32(ewrt);
1207             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1208             ewitab           = _mm_slli_epi32(ewitab,2);
1209             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1210             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1211             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1212             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1213             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1214             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1215             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1216             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1217             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1218
1219             d                = _mm_sub_ps(r10,rswitch);
1220             d                = _mm_max_ps(d,_mm_setzero_ps());
1221             d2               = _mm_mul_ps(d,d);
1222             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1223
1224             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1225
1226             /* Evaluate switch function */
1227             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1228             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1229             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1230
1231             fscal            = felec;
1232
1233             fscal            = _mm_and_ps(fscal,cutoff_mask);
1234
1235             /* Calculate temporary vectorial force */
1236             tx               = _mm_mul_ps(fscal,dx10);
1237             ty               = _mm_mul_ps(fscal,dy10);
1238             tz               = _mm_mul_ps(fscal,dz10);
1239
1240             /* Update vectorial force */
1241             fix1             = _mm_add_ps(fix1,tx);
1242             fiy1             = _mm_add_ps(fiy1,ty);
1243             fiz1             = _mm_add_ps(fiz1,tz);
1244
1245             fjx0             = _mm_add_ps(fjx0,tx);
1246             fjy0             = _mm_add_ps(fjy0,ty);
1247             fjz0             = _mm_add_ps(fjz0,tz);
1248
1249             }
1250
1251             /**************************
1252              * CALCULATE INTERACTIONS *
1253              **************************/
1254
1255             if (gmx_mm_any_lt(rsq20,rcutoff2))
1256             {
1257
1258             r20              = _mm_mul_ps(rsq20,rinv20);
1259
1260             /* Compute parameters for interactions between i and j atoms */
1261             qq20             = _mm_mul_ps(iq2,jq0);
1262
1263             /* EWALD ELECTROSTATICS */
1264
1265             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1266             ewrt             = _mm_mul_ps(r20,ewtabscale);
1267             ewitab           = _mm_cvttps_epi32(ewrt);
1268             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1269             ewitab           = _mm_slli_epi32(ewitab,2);
1270             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1271             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1272             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1273             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1274             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1275             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1276             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1277             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1278             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1279
1280             d                = _mm_sub_ps(r20,rswitch);
1281             d                = _mm_max_ps(d,_mm_setzero_ps());
1282             d2               = _mm_mul_ps(d,d);
1283             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1284
1285             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1286
1287             /* Evaluate switch function */
1288             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1289             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1290             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1291
1292             fscal            = felec;
1293
1294             fscal            = _mm_and_ps(fscal,cutoff_mask);
1295
1296             /* Calculate temporary vectorial force */
1297             tx               = _mm_mul_ps(fscal,dx20);
1298             ty               = _mm_mul_ps(fscal,dy20);
1299             tz               = _mm_mul_ps(fscal,dz20);
1300
1301             /* Update vectorial force */
1302             fix2             = _mm_add_ps(fix2,tx);
1303             fiy2             = _mm_add_ps(fiy2,ty);
1304             fiz2             = _mm_add_ps(fiz2,tz);
1305
1306             fjx0             = _mm_add_ps(fjx0,tx);
1307             fjy0             = _mm_add_ps(fjy0,ty);
1308             fjz0             = _mm_add_ps(fjz0,tz);
1309
1310             }
1311
1312             /**************************
1313              * CALCULATE INTERACTIONS *
1314              **************************/
1315
1316             if (gmx_mm_any_lt(rsq30,rcutoff2))
1317             {
1318
1319             r30              = _mm_mul_ps(rsq30,rinv30);
1320
1321             /* Compute parameters for interactions between i and j atoms */
1322             qq30             = _mm_mul_ps(iq3,jq0);
1323
1324             /* EWALD ELECTROSTATICS */
1325
1326             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1327             ewrt             = _mm_mul_ps(r30,ewtabscale);
1328             ewitab           = _mm_cvttps_epi32(ewrt);
1329             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1330             ewitab           = _mm_slli_epi32(ewitab,2);
1331             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1332             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1333             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1334             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1335             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1336             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1337             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1338             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
1339             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1340
1341             d                = _mm_sub_ps(r30,rswitch);
1342             d                = _mm_max_ps(d,_mm_setzero_ps());
1343             d2               = _mm_mul_ps(d,d);
1344             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1345
1346             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1347
1348             /* Evaluate switch function */
1349             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1350             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1351             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1352
1353             fscal            = felec;
1354
1355             fscal            = _mm_and_ps(fscal,cutoff_mask);
1356
1357             /* Calculate temporary vectorial force */
1358             tx               = _mm_mul_ps(fscal,dx30);
1359             ty               = _mm_mul_ps(fscal,dy30);
1360             tz               = _mm_mul_ps(fscal,dz30);
1361
1362             /* Update vectorial force */
1363             fix3             = _mm_add_ps(fix3,tx);
1364             fiy3             = _mm_add_ps(fiy3,ty);
1365             fiz3             = _mm_add_ps(fiz3,tz);
1366
1367             fjx0             = _mm_add_ps(fjx0,tx);
1368             fjy0             = _mm_add_ps(fjy0,ty);
1369             fjz0             = _mm_add_ps(fjz0,tz);
1370
1371             }
1372
1373             fjptrA             = f+j_coord_offsetA;
1374             fjptrB             = f+j_coord_offsetB;
1375             fjptrC             = f+j_coord_offsetC;
1376             fjptrD             = f+j_coord_offsetD;
1377
1378             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1379
1380             /* Inner loop uses 242 flops */
1381         }
1382
1383         if(jidx<j_index_end)
1384         {
1385
1386             /* Get j neighbor index, and coordinate index */
1387             jnrlistA         = jjnr[jidx];
1388             jnrlistB         = jjnr[jidx+1];
1389             jnrlistC         = jjnr[jidx+2];
1390             jnrlistD         = jjnr[jidx+3];
1391             /* Sign of each element will be negative for non-real atoms.
1392              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1393              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1394              */
1395             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1396             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1397             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1398             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1399             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1400             j_coord_offsetA  = DIM*jnrA;
1401             j_coord_offsetB  = DIM*jnrB;
1402             j_coord_offsetC  = DIM*jnrC;
1403             j_coord_offsetD  = DIM*jnrD;
1404
1405             /* load j atom coordinates */
1406             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1407                                               x+j_coord_offsetC,x+j_coord_offsetD,
1408                                               &jx0,&jy0,&jz0);
1409
1410             /* Calculate displacement vector */
1411             dx00             = _mm_sub_ps(ix0,jx0);
1412             dy00             = _mm_sub_ps(iy0,jy0);
1413             dz00             = _mm_sub_ps(iz0,jz0);
1414             dx10             = _mm_sub_ps(ix1,jx0);
1415             dy10             = _mm_sub_ps(iy1,jy0);
1416             dz10             = _mm_sub_ps(iz1,jz0);
1417             dx20             = _mm_sub_ps(ix2,jx0);
1418             dy20             = _mm_sub_ps(iy2,jy0);
1419             dz20             = _mm_sub_ps(iz2,jz0);
1420             dx30             = _mm_sub_ps(ix3,jx0);
1421             dy30             = _mm_sub_ps(iy3,jy0);
1422             dz30             = _mm_sub_ps(iz3,jz0);
1423
1424             /* Calculate squared distance and things based on it */
1425             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1426             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1427             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1428             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1429
1430             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1431             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1432             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1433             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1434
1435             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1436             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1437             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1438             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1439
1440             /* Load parameters for j particles */
1441             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1442                                                               charge+jnrC+0,charge+jnrD+0);
1443             vdwjidx0A        = 2*vdwtype[jnrA+0];
1444             vdwjidx0B        = 2*vdwtype[jnrB+0];
1445             vdwjidx0C        = 2*vdwtype[jnrC+0];
1446             vdwjidx0D        = 2*vdwtype[jnrD+0];
1447
1448             fjx0             = _mm_setzero_ps();
1449             fjy0             = _mm_setzero_ps();
1450             fjz0             = _mm_setzero_ps();
1451
1452             /**************************
1453              * CALCULATE INTERACTIONS *
1454              **************************/
1455
1456             if (gmx_mm_any_lt(rsq00,rcutoff2))
1457             {
1458
1459             r00              = _mm_mul_ps(rsq00,rinv00);
1460             r00              = _mm_andnot_ps(dummy_mask,r00);
1461
1462             /* Compute parameters for interactions between i and j atoms */
1463             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1464                                          vdwparam+vdwioffset0+vdwjidx0B,
1465                                          vdwparam+vdwioffset0+vdwjidx0C,
1466                                          vdwparam+vdwioffset0+vdwjidx0D,
1467                                          &c6_00,&c12_00);
1468
1469             /* LENNARD-JONES DISPERSION/REPULSION */
1470
1471             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1472             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1473             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1474             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1475             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1476
1477             d                = _mm_sub_ps(r00,rswitch);
1478             d                = _mm_max_ps(d,_mm_setzero_ps());
1479             d2               = _mm_mul_ps(d,d);
1480             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1481
1482             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1483
1484             /* Evaluate switch function */
1485             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1486             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1487             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1488
1489             fscal            = fvdw;
1490
1491             fscal            = _mm_and_ps(fscal,cutoff_mask);
1492
1493             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1494
1495             /* Calculate temporary vectorial force */
1496             tx               = _mm_mul_ps(fscal,dx00);
1497             ty               = _mm_mul_ps(fscal,dy00);
1498             tz               = _mm_mul_ps(fscal,dz00);
1499
1500             /* Update vectorial force */
1501             fix0             = _mm_add_ps(fix0,tx);
1502             fiy0             = _mm_add_ps(fiy0,ty);
1503             fiz0             = _mm_add_ps(fiz0,tz);
1504
1505             fjx0             = _mm_add_ps(fjx0,tx);
1506             fjy0             = _mm_add_ps(fjy0,ty);
1507             fjz0             = _mm_add_ps(fjz0,tz);
1508
1509             }
1510
1511             /**************************
1512              * CALCULATE INTERACTIONS *
1513              **************************/
1514
1515             if (gmx_mm_any_lt(rsq10,rcutoff2))
1516             {
1517
1518             r10              = _mm_mul_ps(rsq10,rinv10);
1519             r10              = _mm_andnot_ps(dummy_mask,r10);
1520
1521             /* Compute parameters for interactions between i and j atoms */
1522             qq10             = _mm_mul_ps(iq1,jq0);
1523
1524             /* EWALD ELECTROSTATICS */
1525
1526             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1527             ewrt             = _mm_mul_ps(r10,ewtabscale);
1528             ewitab           = _mm_cvttps_epi32(ewrt);
1529             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1530             ewitab           = _mm_slli_epi32(ewitab,2);
1531             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1532             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1533             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1534             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1535             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1536             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1537             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1538             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1539             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1540
1541             d                = _mm_sub_ps(r10,rswitch);
1542             d                = _mm_max_ps(d,_mm_setzero_ps());
1543             d2               = _mm_mul_ps(d,d);
1544             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1545
1546             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1547
1548             /* Evaluate switch function */
1549             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1550             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1551             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1552
1553             fscal            = felec;
1554
1555             fscal            = _mm_and_ps(fscal,cutoff_mask);
1556
1557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1558
1559             /* Calculate temporary vectorial force */
1560             tx               = _mm_mul_ps(fscal,dx10);
1561             ty               = _mm_mul_ps(fscal,dy10);
1562             tz               = _mm_mul_ps(fscal,dz10);
1563
1564             /* Update vectorial force */
1565             fix1             = _mm_add_ps(fix1,tx);
1566             fiy1             = _mm_add_ps(fiy1,ty);
1567             fiz1             = _mm_add_ps(fiz1,tz);
1568
1569             fjx0             = _mm_add_ps(fjx0,tx);
1570             fjy0             = _mm_add_ps(fjy0,ty);
1571             fjz0             = _mm_add_ps(fjz0,tz);
1572
1573             }
1574
1575             /**************************
1576              * CALCULATE INTERACTIONS *
1577              **************************/
1578
1579             if (gmx_mm_any_lt(rsq20,rcutoff2))
1580             {
1581
1582             r20              = _mm_mul_ps(rsq20,rinv20);
1583             r20              = _mm_andnot_ps(dummy_mask,r20);
1584
1585             /* Compute parameters for interactions between i and j atoms */
1586             qq20             = _mm_mul_ps(iq2,jq0);
1587
1588             /* EWALD ELECTROSTATICS */
1589
1590             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1591             ewrt             = _mm_mul_ps(r20,ewtabscale);
1592             ewitab           = _mm_cvttps_epi32(ewrt);
1593             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1594             ewitab           = _mm_slli_epi32(ewitab,2);
1595             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1596             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1597             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1598             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1599             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1600             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1601             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1602             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1603             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1604
1605             d                = _mm_sub_ps(r20,rswitch);
1606             d                = _mm_max_ps(d,_mm_setzero_ps());
1607             d2               = _mm_mul_ps(d,d);
1608             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1609
1610             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1611
1612             /* Evaluate switch function */
1613             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1614             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1615             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1616
1617             fscal            = felec;
1618
1619             fscal            = _mm_and_ps(fscal,cutoff_mask);
1620
1621             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1622
1623             /* Calculate temporary vectorial force */
1624             tx               = _mm_mul_ps(fscal,dx20);
1625             ty               = _mm_mul_ps(fscal,dy20);
1626             tz               = _mm_mul_ps(fscal,dz20);
1627
1628             /* Update vectorial force */
1629             fix2             = _mm_add_ps(fix2,tx);
1630             fiy2             = _mm_add_ps(fiy2,ty);
1631             fiz2             = _mm_add_ps(fiz2,tz);
1632
1633             fjx0             = _mm_add_ps(fjx0,tx);
1634             fjy0             = _mm_add_ps(fjy0,ty);
1635             fjz0             = _mm_add_ps(fjz0,tz);
1636
1637             }
1638
1639             /**************************
1640              * CALCULATE INTERACTIONS *
1641              **************************/
1642
1643             if (gmx_mm_any_lt(rsq30,rcutoff2))
1644             {
1645
1646             r30              = _mm_mul_ps(rsq30,rinv30);
1647             r30              = _mm_andnot_ps(dummy_mask,r30);
1648
1649             /* Compute parameters for interactions between i and j atoms */
1650             qq30             = _mm_mul_ps(iq3,jq0);
1651
1652             /* EWALD ELECTROSTATICS */
1653
1654             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1655             ewrt             = _mm_mul_ps(r30,ewtabscale);
1656             ewitab           = _mm_cvttps_epi32(ewrt);
1657             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1658             ewitab           = _mm_slli_epi32(ewitab,2);
1659             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1660             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1661             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1662             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1663             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1664             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1665             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1666             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
1667             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1668
1669             d                = _mm_sub_ps(r30,rswitch);
1670             d                = _mm_max_ps(d,_mm_setzero_ps());
1671             d2               = _mm_mul_ps(d,d);
1672             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1673
1674             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1675
1676             /* Evaluate switch function */
1677             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1678             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1679             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1680
1681             fscal            = felec;
1682
1683             fscal            = _mm_and_ps(fscal,cutoff_mask);
1684
1685             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1686
1687             /* Calculate temporary vectorial force */
1688             tx               = _mm_mul_ps(fscal,dx30);
1689             ty               = _mm_mul_ps(fscal,dy30);
1690             tz               = _mm_mul_ps(fscal,dz30);
1691
1692             /* Update vectorial force */
1693             fix3             = _mm_add_ps(fix3,tx);
1694             fiy3             = _mm_add_ps(fiy3,ty);
1695             fiz3             = _mm_add_ps(fiz3,tz);
1696
1697             fjx0             = _mm_add_ps(fjx0,tx);
1698             fjy0             = _mm_add_ps(fjy0,ty);
1699             fjz0             = _mm_add_ps(fjz0,tz);
1700
1701             }
1702
1703             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1704             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1705             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1706             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1707
1708             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1709
1710             /* Inner loop uses 246 flops */
1711         }
1712
1713         /* End of innermost loop */
1714
1715         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1716                                               f+i_coord_offset,fshift+i_shift_offset);
1717
1718         /* Increment number of inner iterations */
1719         inneriter                  += j_index_end - j_index_start;
1720
1721         /* Outer loop uses 24 flops */
1722     }
1723
1724     /* Increment number of outer iterations */
1725     outeriter        += nri;
1726
1727     /* Update outer/inner flops */
1728
1729     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*246);
1730 }