src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecEwSw_VdwLJSw_GeomW3P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_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              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           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     real             *ewtab;
 107     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 108     real             rswitch_scalar,d_scalar;
 109     __m128           dummy_mask,cutoff_mask;
 110     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 111     __m128           one     = _mm_set1_ps(1.0);
 112     __m128           two     = _mm_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_ps(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 131     ewtab            = fr->ic->tabq_coul_FDV0;
 132     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 133     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+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     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 143     rcutoff_scalar   = fr->rcoulomb;
 144     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 145     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 146
 147     rswitch_scalar   = fr->rcoulomb_switch;
 148     rswitch          = _mm_set1_ps(rswitch_scalar);
 149     /* Setup switch parameters */
 150     d_scalar         = rcutoff_scalar-rswitch_scalar;
 151     d                = _mm_set1_ps(d_scalar);
 152     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 153     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 154     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 155     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 156     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 157     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 158
 159     /* Avoid stupid compiler warnings */
 160     jnrA = jnrB = jnrC = jnrD = 0;
 161     j_coord_offsetA = 0;
 162     j_coord_offsetB = 0;
 163     j_coord_offsetC = 0;
 164     j_coord_offsetD = 0;
 165
 166     outeriter        = 0;
 167     inneriter        = 0;
 168
 169     for(iidx=0;iidx<4*DIM;iidx++)
 170     {
 171         scratch[iidx] = 0.0;
 172     }
 173
 174     /* Start outer loop over neighborlists */
 175     for(iidx=0; iidx<nri; iidx++)
 176     {
 177         /* Load shift vector for this list */
 178         i_shift_offset   = DIM*shiftidx[iidx];
 179
 180         /* Load limits for loop over neighbors */
 181         j_index_start    = jindex[iidx];
 182         j_index_end      = jindex[iidx+1];
 183
 184         /* Get outer coordinate index */
 185         inr              = iinr[iidx];
 186         i_coord_offset   = DIM*inr;
 187
 188         /* Load i particle coords and add shift vector */
 189         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 190                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 191
 192         fix0             = _mm_setzero_ps();
 193         fiy0             = _mm_setzero_ps();
 194         fiz0             = _mm_setzero_ps();
 195         fix1             = _mm_setzero_ps();
 196         fiy1             = _mm_setzero_ps();
 197         fiz1             = _mm_setzero_ps();
 198         fix2             = _mm_setzero_ps();
 199         fiy2             = _mm_setzero_ps();
 200         fiz2             = _mm_setzero_ps();
 201
 202         /* Reset potential sums */
 203         velecsum         = _mm_setzero_ps();
 204         vvdwsum          = _mm_setzero_ps();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             jnrC             = jjnr[jidx+2];
 214             jnrD             = jjnr[jidx+3];
 215             j_coord_offsetA  = DIM*jnrA;
 216             j_coord_offsetB  = DIM*jnrB;
 217             j_coord_offsetC  = DIM*jnrC;
 218             j_coord_offsetD  = DIM*jnrD;
 219
 220             /* load j atom coordinates */
 221             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                               x+j_coord_offsetC,x+j_coord_offsetD,
 223                                               &jx0,&jy0,&jz0);
 224
 225             /* Calculate displacement vector */
 226             dx00             = _mm_sub_ps(ix0,jx0);
 227             dy00             = _mm_sub_ps(iy0,jy0);
 228             dz00             = _mm_sub_ps(iz0,jz0);
 229             dx10             = _mm_sub_ps(ix1,jx0);
 230             dy10             = _mm_sub_ps(iy1,jy0);
 231             dz10             = _mm_sub_ps(iz1,jz0);
 232             dx20             = _mm_sub_ps(ix2,jx0);
 233             dy20             = _mm_sub_ps(iy2,jy0);
 234             dz20             = _mm_sub_ps(iz2,jz0);
 235
 236             /* Calculate squared distance and things based on it */
 237             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 238             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 239             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 240
 241             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 242             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 243             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 244
 245             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 246             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 247             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 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             r00              = _mm_mul_ps(rsq00,rinv00);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq00             = _mm_mul_ps(iq0,jq0);
 272             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 273                                          vdwparam+vdwioffset0+vdwjidx0B,
 274                                          vdwparam+vdwioffset0+vdwjidx0C,
 275                                          vdwparam+vdwioffset0+vdwjidx0D,
 276                                          &c6_00,&c12_00);
 277
 278             /* EWALD ELECTROSTATICS */
 279
 280             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 281             ewrt             = _mm_mul_ps(r00,ewtabscale);
 282             ewitab           = _mm_cvttps_epi32(ewrt);
 283             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 284             ewitab           = _mm_slli_epi32(ewitab,2);
 285             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 286             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 287             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 288             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 289             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 290             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 291             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 292             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 293             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 294
 295             /* LENNARD-JONES DISPERSION/REPULSION */
 296
 297             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 298             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 299             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 300             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 301             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 302
 303             d                = _mm_sub_ps(r00,rswitch);
 304             d                = _mm_max_ps(d,_mm_setzero_ps());
 305             d2               = _mm_mul_ps(d,d);
 306             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)))))));
 307
 308             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 309
 310             /* Evaluate switch function */
 311             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 312             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 313             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 314             velec            = _mm_mul_ps(velec,sw);
 315             vvdw             = _mm_mul_ps(vvdw,sw);
 316             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velec            = _mm_and_ps(velec,cutoff_mask);
 320             velecsum         = _mm_add_ps(velecsum,velec);
 321             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 322             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 323
 324             fscal            = _mm_add_ps(felec,fvdw);
 325
 326             fscal            = _mm_and_ps(fscal,cutoff_mask);
 327
 328             /* Calculate temporary vectorial force */
 329             tx               = _mm_mul_ps(fscal,dx00);
 330             ty               = _mm_mul_ps(fscal,dy00);
 331             tz               = _mm_mul_ps(fscal,dz00);
 332
 333             /* Update vectorial force */
 334             fix0             = _mm_add_ps(fix0,tx);
 335             fiy0             = _mm_add_ps(fiy0,ty);
 336             fiz0             = _mm_add_ps(fiz0,tz);
 337
 338             fjx0             = _mm_add_ps(fjx0,tx);
 339             fjy0             = _mm_add_ps(fjy0,ty);
 340             fjz0             = _mm_add_ps(fjz0,tz);
 341
 342             }
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             if (gmx_mm_any_lt(rsq10,rcutoff2))
 349             {
 350
 351             r10              = _mm_mul_ps(rsq10,rinv10);
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq10             = _mm_mul_ps(iq1,jq0);
 355
 356             /* EWALD ELECTROSTATICS */
 357
 358             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 359             ewrt             = _mm_mul_ps(r10,ewtabscale);
 360             ewitab           = _mm_cvttps_epi32(ewrt);
 361             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 362             ewitab           = _mm_slli_epi32(ewitab,2);
 363             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 364             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 365             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 366             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 367             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 368             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 369             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 370             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 371             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 372
 373             d                = _mm_sub_ps(r10,rswitch);
 374             d                = _mm_max_ps(d,_mm_setzero_ps());
 375             d2               = _mm_mul_ps(d,d);
 376             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)))))));
 377
 378             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 379
 380             /* Evaluate switch function */
 381             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 382             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 383             velec            = _mm_mul_ps(velec,sw);
 384             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm_and_ps(velec,cutoff_mask);
 388             velecsum         = _mm_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             fscal            = _mm_and_ps(fscal,cutoff_mask);
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm_mul_ps(fscal,dx10);
 396             ty               = _mm_mul_ps(fscal,dy10);
 397             tz               = _mm_mul_ps(fscal,dz10);
 398
 399             /* Update vectorial force */
 400             fix1             = _mm_add_ps(fix1,tx);
 401             fiy1             = _mm_add_ps(fiy1,ty);
 402             fiz1             = _mm_add_ps(fiz1,tz);
 403
 404             fjx0             = _mm_add_ps(fjx0,tx);
 405             fjy0             = _mm_add_ps(fjy0,ty);
 406             fjz0             = _mm_add_ps(fjz0,tz);
 407
 408             }
 409
 410             /**************************
 411              * CALCULATE INTERACTIONS *
 412              **************************/
 413
 414             if (gmx_mm_any_lt(rsq20,rcutoff2))
 415             {
 416
 417             r20              = _mm_mul_ps(rsq20,rinv20);
 418
 419             /* Compute parameters for interactions between i and j atoms */
 420             qq20             = _mm_mul_ps(iq2,jq0);
 421
 422             /* EWALD ELECTROSTATICS */
 423
 424             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 425             ewrt             = _mm_mul_ps(r20,ewtabscale);
 426             ewitab           = _mm_cvttps_epi32(ewrt);
 427             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 428             ewitab           = _mm_slli_epi32(ewitab,2);
 429             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 430             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 431             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 432             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 433             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 434             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 435             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 436             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 437             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 438
 439             d                = _mm_sub_ps(r20,rswitch);
 440             d                = _mm_max_ps(d,_mm_setzero_ps());
 441             d2               = _mm_mul_ps(d,d);
 442             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)))))));
 443
 444             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 445
 446             /* Evaluate switch function */
 447             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 448             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 449             velec            = _mm_mul_ps(velec,sw);
 450             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 451
 452             /* Update potential sum for this i atom from the interaction with this j atom. */
 453             velec            = _mm_and_ps(velec,cutoff_mask);
 454             velecsum         = _mm_add_ps(velecsum,velec);
 455
 456             fscal            = felec;
 457
 458             fscal            = _mm_and_ps(fscal,cutoff_mask);
 459
 460             /* Calculate temporary vectorial force */
 461             tx               = _mm_mul_ps(fscal,dx20);
 462             ty               = _mm_mul_ps(fscal,dy20);
 463             tz               = _mm_mul_ps(fscal,dz20);
 464
 465             /* Update vectorial force */
 466             fix2             = _mm_add_ps(fix2,tx);
 467             fiy2             = _mm_add_ps(fiy2,ty);
 468             fiz2             = _mm_add_ps(fiz2,tz);
 469
 470             fjx0             = _mm_add_ps(fjx0,tx);
 471             fjy0             = _mm_add_ps(fjy0,ty);
 472             fjz0             = _mm_add_ps(fjz0,tz);
 473
 474             }
 475
 476             fjptrA             = f+j_coord_offsetA;
 477             fjptrB             = f+j_coord_offsetB;
 478             fjptrC             = f+j_coord_offsetC;
 479             fjptrD             = f+j_coord_offsetD;
 480
 481             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 482
 483             /* Inner loop uses 213 flops */
 484         }
 485
 486         if(jidx<j_index_end)
 487         {
 488
 489             /* Get j neighbor index, and coordinate index */
 490             jnrlistA         = jjnr[jidx];
 491             jnrlistB         = jjnr[jidx+1];
 492             jnrlistC         = jjnr[jidx+2];
 493             jnrlistD         = jjnr[jidx+3];
 494             /* Sign of each element will be negative for non-real atoms.
 495              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 496              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 497              */
 498             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 499             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 500             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 501             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 502             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 503             j_coord_offsetA  = DIM*jnrA;
 504             j_coord_offsetB  = DIM*jnrB;
 505             j_coord_offsetC  = DIM*jnrC;
 506             j_coord_offsetD  = DIM*jnrD;
 507
 508             /* load j atom coordinates */
 509             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 510                                               x+j_coord_offsetC,x+j_coord_offsetD,
 511                                               &jx0,&jy0,&jz0);
 512
 513             /* Calculate displacement vector */
 514             dx00             = _mm_sub_ps(ix0,jx0);
 515             dy00             = _mm_sub_ps(iy0,jy0);
 516             dz00             = _mm_sub_ps(iz0,jz0);
 517             dx10             = _mm_sub_ps(ix1,jx0);
 518             dy10             = _mm_sub_ps(iy1,jy0);
 519             dz10             = _mm_sub_ps(iz1,jz0);
 520             dx20             = _mm_sub_ps(ix2,jx0);
 521             dy20             = _mm_sub_ps(iy2,jy0);
 522             dz20             = _mm_sub_ps(iz2,jz0);
 523
 524             /* Calculate squared distance and things based on it */
 525             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 526             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 527             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 528
 529             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 530             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 531             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 532
 533             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 534             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 535             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 536
 537             /* Load parameters for j particles */
 538             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 539                                                               charge+jnrC+0,charge+jnrD+0);
 540             vdwjidx0A        = 2*vdwtype[jnrA+0];
 541             vdwjidx0B        = 2*vdwtype[jnrB+0];
 542             vdwjidx0C        = 2*vdwtype[jnrC+0];
 543             vdwjidx0D        = 2*vdwtype[jnrD+0];
 544
 545             fjx0             = _mm_setzero_ps();
 546             fjy0             = _mm_setzero_ps();
 547             fjz0             = _mm_setzero_ps();
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             if (gmx_mm_any_lt(rsq00,rcutoff2))
 554             {
 555
 556             r00              = _mm_mul_ps(rsq00,rinv00);
 557             r00              = _mm_andnot_ps(dummy_mask,r00);
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             qq00             = _mm_mul_ps(iq0,jq0);
 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             /* EWALD ELECTROSTATICS */
 568
 569             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 570             ewrt             = _mm_mul_ps(r00,ewtabscale);
 571             ewitab           = _mm_cvttps_epi32(ewrt);
 572             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 573             ewitab           = _mm_slli_epi32(ewitab,2);
 574             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 575             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 576             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 577             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 578             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 579             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 580             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 581             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 582             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 583
 584             /* LENNARD-JONES DISPERSION/REPULSION */
 585
 586             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 587             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 588             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 589             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 590             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 591
 592             d                = _mm_sub_ps(r00,rswitch);
 593             d                = _mm_max_ps(d,_mm_setzero_ps());
 594             d2               = _mm_mul_ps(d,d);
 595             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)))))));
 596
 597             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 598
 599             /* Evaluate switch function */
 600             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 601             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 602             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 603             velec            = _mm_mul_ps(velec,sw);
 604             vvdw             = _mm_mul_ps(vvdw,sw);
 605             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velec            = _mm_and_ps(velec,cutoff_mask);
 609             velec            = _mm_andnot_ps(dummy_mask,velec);
 610             velecsum         = _mm_add_ps(velecsum,velec);
 611             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 612             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 613             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 614
 615             fscal            = _mm_add_ps(felec,fvdw);
 616
 617             fscal            = _mm_and_ps(fscal,cutoff_mask);
 618
 619             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 620
 621             /* Calculate temporary vectorial force */
 622             tx               = _mm_mul_ps(fscal,dx00);
 623             ty               = _mm_mul_ps(fscal,dy00);
 624             tz               = _mm_mul_ps(fscal,dz00);
 625
 626             /* Update vectorial force */
 627             fix0             = _mm_add_ps(fix0,tx);
 628             fiy0             = _mm_add_ps(fiy0,ty);
 629             fiz0             = _mm_add_ps(fiz0,tz);
 630
 631             fjx0             = _mm_add_ps(fjx0,tx);
 632             fjy0             = _mm_add_ps(fjy0,ty);
 633             fjz0             = _mm_add_ps(fjz0,tz);
 634
 635             }
 636
 637             /**************************
 638              * CALCULATE INTERACTIONS *
 639              **************************/
 640
 641             if (gmx_mm_any_lt(rsq10,rcutoff2))
 642             {
 643
 644             r10              = _mm_mul_ps(rsq10,rinv10);
 645             r10              = _mm_andnot_ps(dummy_mask,r10);
 646
 647             /* Compute parameters for interactions between i and j atoms */
 648             qq10             = _mm_mul_ps(iq1,jq0);
 649
 650             /* EWALD ELECTROSTATICS */
 651
 652             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 653             ewrt             = _mm_mul_ps(r10,ewtabscale);
 654             ewitab           = _mm_cvttps_epi32(ewrt);
 655             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 656             ewitab           = _mm_slli_epi32(ewitab,2);
 657             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 658             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 659             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 660             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 661             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 662             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 663             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 664             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 665             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 666
 667             d                = _mm_sub_ps(r10,rswitch);
 668             d                = _mm_max_ps(d,_mm_setzero_ps());
 669             d2               = _mm_mul_ps(d,d);
 670             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)))))));
 671
 672             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 673
 674             /* Evaluate switch function */
 675             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 676             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 677             velec            = _mm_mul_ps(velec,sw);
 678             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 679
 680             /* Update potential sum for this i atom from the interaction with this j atom. */
 681             velec            = _mm_and_ps(velec,cutoff_mask);
 682             velec            = _mm_andnot_ps(dummy_mask,velec);
 683             velecsum         = _mm_add_ps(velecsum,velec);
 684
 685             fscal            = felec;
 686
 687             fscal            = _mm_and_ps(fscal,cutoff_mask);
 688
 689             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 690
 691             /* Calculate temporary vectorial force */
 692             tx               = _mm_mul_ps(fscal,dx10);
 693             ty               = _mm_mul_ps(fscal,dy10);
 694             tz               = _mm_mul_ps(fscal,dz10);
 695
 696             /* Update vectorial force */
 697             fix1             = _mm_add_ps(fix1,tx);
 698             fiy1             = _mm_add_ps(fiy1,ty);
 699             fiz1             = _mm_add_ps(fiz1,tz);
 700
 701             fjx0             = _mm_add_ps(fjx0,tx);
 702             fjy0             = _mm_add_ps(fjy0,ty);
 703             fjz0             = _mm_add_ps(fjz0,tz);
 704
 705             }
 706
 707             /**************************
 708              * CALCULATE INTERACTIONS *
 709              **************************/
 710
 711             if (gmx_mm_any_lt(rsq20,rcutoff2))
 712             {
 713
 714             r20              = _mm_mul_ps(rsq20,rinv20);
 715             r20              = _mm_andnot_ps(dummy_mask,r20);
 716
 717             /* Compute parameters for interactions between i and j atoms */
 718             qq20             = _mm_mul_ps(iq2,jq0);
 719
 720             /* EWALD ELECTROSTATICS */
 721
 722             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 723             ewrt             = _mm_mul_ps(r20,ewtabscale);
 724             ewitab           = _mm_cvttps_epi32(ewrt);
 725             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 726             ewitab           = _mm_slli_epi32(ewitab,2);
 727             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 728             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 729             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 730             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 731             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 732             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 733             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 734             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 735             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 736
 737             d                = _mm_sub_ps(r20,rswitch);
 738             d                = _mm_max_ps(d,_mm_setzero_ps());
 739             d2               = _mm_mul_ps(d,d);
 740             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)))))));
 741
 742             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 743
 744             /* Evaluate switch function */
 745             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 746             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 747             velec            = _mm_mul_ps(velec,sw);
 748             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 749
 750             /* Update potential sum for this i atom from the interaction with this j atom. */
 751             velec            = _mm_and_ps(velec,cutoff_mask);
 752             velec            = _mm_andnot_ps(dummy_mask,velec);
 753             velecsum         = _mm_add_ps(velecsum,velec);
 754
 755             fscal            = felec;
 756
 757             fscal            = _mm_and_ps(fscal,cutoff_mask);
 758
 759             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 760
 761             /* Calculate temporary vectorial force */
 762             tx               = _mm_mul_ps(fscal,dx20);
 763             ty               = _mm_mul_ps(fscal,dy20);
 764             tz               = _mm_mul_ps(fscal,dz20);
 765
 766             /* Update vectorial force */
 767             fix2             = _mm_add_ps(fix2,tx);
 768             fiy2             = _mm_add_ps(fiy2,ty);
 769             fiz2             = _mm_add_ps(fiz2,tz);
 770
 771             fjx0             = _mm_add_ps(fjx0,tx);
 772             fjy0             = _mm_add_ps(fjy0,ty);
 773             fjz0             = _mm_add_ps(fjz0,tz);
 774
 775             }
 776
 777             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 778             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 779             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 780             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 781
 782             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 783
 784             /* Inner loop uses 216 flops */
 785         }
 786
 787         /* End of innermost loop */
 788
 789         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 790                                               f+i_coord_offset,fshift+i_shift_offset);
 791
 792         ggid                        = gid[iidx];
 793         /* Update potential energies */
 794         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 795         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 796
 797         /* Increment number of inner iterations */
 798         inneriter                  += j_index_end - j_index_start;
 799
 800         /* Outer loop uses 20 flops */
 801     }
 802
 803     /* Increment number of outer iterations */
 804     outeriter        += nri;
 805
 806     /* Update outer/inner flops */
 807
 808     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*216);
 809 }
 810 /*
 811  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_sse4_1_single
 812  * Electrostatics interaction: Ewald
 813  * VdW interaction:            LennardJones
 814  * Geometry:                   Water3-Particle
 815  * Calculate force/pot:        Force
 816  */
 817 void
 818 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_sse4_1_single
 819                     (t_nblist                    * gmx_restrict       nlist,
 820                      rvec                        * gmx_restrict          xx,
 821                      rvec                        * gmx_restrict          ff,
 822                      t_forcerec                  * gmx_restrict          fr,
 823                      t_mdatoms                   * gmx_restrict     mdatoms,
 824                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 825                      t_nrnb                      * gmx_restrict        nrnb)
 826 {
 827     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 828      * just 0 for non-waters.
 829      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 830      * jnr indices corresponding to data put in the four positions in the SIMD register.
 831      */
 832     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 833     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 834     int              jnrA,jnrB,jnrC,jnrD;
 835     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 836     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 837     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 838     real             rcutoff_scalar;
 839     real             *shiftvec,*fshift,*x,*f;
 840     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 841     real             scratch[4*DIM];
 842     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 843     int              vdwioffset0;
 844     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 845     int              vdwioffset1;
 846     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 847     int              vdwioffset2;
 848     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 849     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 850     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 851     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 852     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 853     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 854     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 855     real             *charge;
 856     int              nvdwtype;
 857     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 858     int              *vdwtype;
 859     real             *vdwparam;
 860     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 861     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 862     __m128i          ewitab;
 863     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 864     real             *ewtab;
 865     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 866     real             rswitch_scalar,d_scalar;
 867     __m128           dummy_mask,cutoff_mask;
 868     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 869     __m128           one     = _mm_set1_ps(1.0);
 870     __m128           two     = _mm_set1_ps(2.0);
 871     x                = xx[0];
 872     f                = ff[0];
 873
 874     nri              = nlist->nri;
 875     iinr             = nlist->iinr;
 876     jindex           = nlist->jindex;
 877     jjnr             = nlist->jjnr;
 878     shiftidx         = nlist->shift;
 879     gid              = nlist->gid;
 880     shiftvec         = fr->shift_vec[0];
 881     fshift           = fr->fshift[0];
 882     facel            = _mm_set1_ps(fr->epsfac);
 883     charge           = mdatoms->chargeA;
 884     nvdwtype         = fr->ntype;
 885     vdwparam         = fr->nbfp;
 886     vdwtype          = mdatoms->typeA;
 887
 888     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 889     ewtab            = fr->ic->tabq_coul_FDV0;
 890     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 891     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 892
 893     /* Setup water-specific parameters */
 894     inr              = nlist->iinr[0];
 895     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 896     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 897     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 898     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 899
 900     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 901     rcutoff_scalar   = fr->rcoulomb;
 902     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 903     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 904
 905     rswitch_scalar   = fr->rcoulomb_switch;
 906     rswitch          = _mm_set1_ps(rswitch_scalar);
 907     /* Setup switch parameters */
 908     d_scalar         = rcutoff_scalar-rswitch_scalar;
 909     d                = _mm_set1_ps(d_scalar);
 910     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 911     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 912     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 913     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 914     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 915     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 916
 917     /* Avoid stupid compiler warnings */
 918     jnrA = jnrB = jnrC = jnrD = 0;
 919     j_coord_offsetA = 0;
 920     j_coord_offsetB = 0;
 921     j_coord_offsetC = 0;
 922     j_coord_offsetD = 0;
 923
 924     outeriter        = 0;
 925     inneriter        = 0;
 926
 927     for(iidx=0;iidx<4*DIM;iidx++)
 928     {
 929         scratch[iidx] = 0.0;
 930     }
 931
 932     /* Start outer loop over neighborlists */
 933     for(iidx=0; iidx<nri; iidx++)
 934     {
 935         /* Load shift vector for this list */
 936         i_shift_offset   = DIM*shiftidx[iidx];
 937
 938         /* Load limits for loop over neighbors */
 939         j_index_start    = jindex[iidx];
 940         j_index_end      = jindex[iidx+1];
 941
 942         /* Get outer coordinate index */
 943         inr              = iinr[iidx];
 944         i_coord_offset   = DIM*inr;
 945
 946         /* Load i particle coords and add shift vector */
 947         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 948                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 949
 950         fix0             = _mm_setzero_ps();
 951         fiy0             = _mm_setzero_ps();
 952         fiz0             = _mm_setzero_ps();
 953         fix1             = _mm_setzero_ps();
 954         fiy1             = _mm_setzero_ps();
 955         fiz1             = _mm_setzero_ps();
 956         fix2             = _mm_setzero_ps();
 957         fiy2             = _mm_setzero_ps();
 958         fiz2             = _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
 990             /* Calculate squared distance and things based on it */
 991             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 992             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 993             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 994
 995             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 996             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 997             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 998
 999             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1000             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1001             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1002
1003             /* Load parameters for j particles */
1004             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1005                                                               charge+jnrC+0,charge+jnrD+0);
1006             vdwjidx0A        = 2*vdwtype[jnrA+0];
1007             vdwjidx0B        = 2*vdwtype[jnrB+0];
1008             vdwjidx0C        = 2*vdwtype[jnrC+0];
1009             vdwjidx0D        = 2*vdwtype[jnrD+0];
1010
1011             fjx0             = _mm_setzero_ps();
1012             fjy0             = _mm_setzero_ps();
1013             fjz0             = _mm_setzero_ps();
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             if (gmx_mm_any_lt(rsq00,rcutoff2))
1020             {
1021
1022             r00              = _mm_mul_ps(rsq00,rinv00);
1023
1024             /* Compute parameters for interactions between i and j atoms */
1025             qq00             = _mm_mul_ps(iq0,jq0);
1026             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1027                                          vdwparam+vdwioffset0+vdwjidx0B,
1028                                          vdwparam+vdwioffset0+vdwjidx0C,
1029                                          vdwparam+vdwioffset0+vdwjidx0D,
1030                                          &c6_00,&c12_00);
1031
1032             /* EWALD ELECTROSTATICS */
1033
1034             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1035             ewrt             = _mm_mul_ps(r00,ewtabscale);
1036             ewitab           = _mm_cvttps_epi32(ewrt);
1037             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1038             ewitab           = _mm_slli_epi32(ewitab,2);
1039             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1040             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1041             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1042             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1043             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1044             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1045             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1046             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
1047             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1048
1049             /* LENNARD-JONES DISPERSION/REPULSION */
1050
1051             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1052             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1053             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1054             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1055             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1056
1057             d                = _mm_sub_ps(r00,rswitch);
1058             d                = _mm_max_ps(d,_mm_setzero_ps());
1059             d2               = _mm_mul_ps(d,d);
1060             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)))))));
1061
1062             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1063
1064             /* Evaluate switch function */
1065             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1066             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1067             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1068             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1069
1070             fscal            = _mm_add_ps(felec,fvdw);
1071
1072             fscal            = _mm_and_ps(fscal,cutoff_mask);
1073
1074             /* Calculate temporary vectorial force */
1075             tx               = _mm_mul_ps(fscal,dx00);
1076             ty               = _mm_mul_ps(fscal,dy00);
1077             tz               = _mm_mul_ps(fscal,dz00);
1078
1079             /* Update vectorial force */
1080             fix0             = _mm_add_ps(fix0,tx);
1081             fiy0             = _mm_add_ps(fiy0,ty);
1082             fiz0             = _mm_add_ps(fiz0,tz);
1083
1084             fjx0             = _mm_add_ps(fjx0,tx);
1085             fjy0             = _mm_add_ps(fjy0,ty);
1086             fjz0             = _mm_add_ps(fjz0,tz);
1087
1088             }
1089
1090             /**************************
1091              * CALCULATE INTERACTIONS *
1092              **************************/
1093
1094             if (gmx_mm_any_lt(rsq10,rcutoff2))
1095             {
1096
1097             r10              = _mm_mul_ps(rsq10,rinv10);
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq10             = _mm_mul_ps(iq1,jq0);
1101
1102             /* EWALD ELECTROSTATICS */
1103
1104             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1105             ewrt             = _mm_mul_ps(r10,ewtabscale);
1106             ewitab           = _mm_cvttps_epi32(ewrt);
1107             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1108             ewitab           = _mm_slli_epi32(ewitab,2);
1109             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1110             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1111             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1112             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1113             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1114             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1115             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1116             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1117             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1118
1119             d                = _mm_sub_ps(r10,rswitch);
1120             d                = _mm_max_ps(d,_mm_setzero_ps());
1121             d2               = _mm_mul_ps(d,d);
1122             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)))))));
1123
1124             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1125
1126             /* Evaluate switch function */
1127             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1128             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1129             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1130
1131             fscal            = felec;
1132
1133             fscal            = _mm_and_ps(fscal,cutoff_mask);
1134
1135             /* Calculate temporary vectorial force */
1136             tx               = _mm_mul_ps(fscal,dx10);
1137             ty               = _mm_mul_ps(fscal,dy10);
1138             tz               = _mm_mul_ps(fscal,dz10);
1139
1140             /* Update vectorial force */
1141             fix1             = _mm_add_ps(fix1,tx);
1142             fiy1             = _mm_add_ps(fiy1,ty);
1143             fiz1             = _mm_add_ps(fiz1,tz);
1144
1145             fjx0             = _mm_add_ps(fjx0,tx);
1146             fjy0             = _mm_add_ps(fjy0,ty);
1147             fjz0             = _mm_add_ps(fjz0,tz);
1148
1149             }
1150
1151             /**************************
1152              * CALCULATE INTERACTIONS *
1153              **************************/
1154
1155             if (gmx_mm_any_lt(rsq20,rcutoff2))
1156             {
1157
1158             r20              = _mm_mul_ps(rsq20,rinv20);
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             qq20             = _mm_mul_ps(iq2,jq0);
1162
1163             /* EWALD ELECTROSTATICS */
1164
1165             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1166             ewrt             = _mm_mul_ps(r20,ewtabscale);
1167             ewitab           = _mm_cvttps_epi32(ewrt);
1168             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1169             ewitab           = _mm_slli_epi32(ewitab,2);
1170             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1171             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1172             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1173             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1174             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1175             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1176             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1177             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1178             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1179
1180             d                = _mm_sub_ps(r20,rswitch);
1181             d                = _mm_max_ps(d,_mm_setzero_ps());
1182             d2               = _mm_mul_ps(d,d);
1183             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)))))));
1184
1185             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1186
1187             /* Evaluate switch function */
1188             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1189             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1190             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1191
1192             fscal            = felec;
1193
1194             fscal            = _mm_and_ps(fscal,cutoff_mask);
1195
1196             /* Calculate temporary vectorial force */
1197             tx               = _mm_mul_ps(fscal,dx20);
1198             ty               = _mm_mul_ps(fscal,dy20);
1199             tz               = _mm_mul_ps(fscal,dz20);
1200
1201             /* Update vectorial force */
1202             fix2             = _mm_add_ps(fix2,tx);
1203             fiy2             = _mm_add_ps(fiy2,ty);
1204             fiz2             = _mm_add_ps(fiz2,tz);
1205
1206             fjx0             = _mm_add_ps(fjx0,tx);
1207             fjy0             = _mm_add_ps(fjy0,ty);
1208             fjz0             = _mm_add_ps(fjz0,tz);
1209
1210             }
1211
1212             fjptrA             = f+j_coord_offsetA;
1213             fjptrB             = f+j_coord_offsetB;
1214             fjptrC             = f+j_coord_offsetC;
1215             fjptrD             = f+j_coord_offsetD;
1216
1217             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1218
1219             /* Inner loop uses 201 flops */
1220         }
1221
1222         if(jidx<j_index_end)
1223         {
1224
1225             /* Get j neighbor index, and coordinate index */
1226             jnrlistA         = jjnr[jidx];
1227             jnrlistB         = jjnr[jidx+1];
1228             jnrlistC         = jjnr[jidx+2];
1229             jnrlistD         = jjnr[jidx+3];
1230             /* Sign of each element will be negative for non-real atoms.
1231              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1232              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1233              */
1234             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1235             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1236             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1237             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1238             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1239             j_coord_offsetA  = DIM*jnrA;
1240             j_coord_offsetB  = DIM*jnrB;
1241             j_coord_offsetC  = DIM*jnrC;
1242             j_coord_offsetD  = DIM*jnrD;
1243
1244             /* load j atom coordinates */
1245             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1246                                               x+j_coord_offsetC,x+j_coord_offsetD,
1247                                               &jx0,&jy0,&jz0);
1248
1249             /* Calculate displacement vector */
1250             dx00             = _mm_sub_ps(ix0,jx0);
1251             dy00             = _mm_sub_ps(iy0,jy0);
1252             dz00             = _mm_sub_ps(iz0,jz0);
1253             dx10             = _mm_sub_ps(ix1,jx0);
1254             dy10             = _mm_sub_ps(iy1,jy0);
1255             dz10             = _mm_sub_ps(iz1,jz0);
1256             dx20             = _mm_sub_ps(ix2,jx0);
1257             dy20             = _mm_sub_ps(iy2,jy0);
1258             dz20             = _mm_sub_ps(iz2,jz0);
1259
1260             /* Calculate squared distance and things based on it */
1261             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1262             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1263             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1264
1265             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1266             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1267             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1268
1269             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1270             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1271             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1272
1273             /* Load parameters for j particles */
1274             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1275                                                               charge+jnrC+0,charge+jnrD+0);
1276             vdwjidx0A        = 2*vdwtype[jnrA+0];
1277             vdwjidx0B        = 2*vdwtype[jnrB+0];
1278             vdwjidx0C        = 2*vdwtype[jnrC+0];
1279             vdwjidx0D        = 2*vdwtype[jnrD+0];
1280
1281             fjx0             = _mm_setzero_ps();
1282             fjy0             = _mm_setzero_ps();
1283             fjz0             = _mm_setzero_ps();
1284
1285             /**************************
1286              * CALCULATE INTERACTIONS *
1287              **************************/
1288
1289             if (gmx_mm_any_lt(rsq00,rcutoff2))
1290             {
1291
1292             r00              = _mm_mul_ps(rsq00,rinv00);
1293             r00              = _mm_andnot_ps(dummy_mask,r00);
1294
1295             /* Compute parameters for interactions between i and j atoms */
1296             qq00             = _mm_mul_ps(iq0,jq0);
1297             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1298                                          vdwparam+vdwioffset0+vdwjidx0B,
1299                                          vdwparam+vdwioffset0+vdwjidx0C,
1300                                          vdwparam+vdwioffset0+vdwjidx0D,
1301                                          &c6_00,&c12_00);
1302
1303             /* EWALD ELECTROSTATICS */
1304
1305             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1306             ewrt             = _mm_mul_ps(r00,ewtabscale);
1307             ewitab           = _mm_cvttps_epi32(ewrt);
1308             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1309             ewitab           = _mm_slli_epi32(ewitab,2);
1310             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1311             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1312             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1313             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1314             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1315             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1316             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1317             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
1318             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1319
1320             /* LENNARD-JONES DISPERSION/REPULSION */
1321
1322             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1323             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1324             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1325             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1326             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1327
1328             d                = _mm_sub_ps(r00,rswitch);
1329             d                = _mm_max_ps(d,_mm_setzero_ps());
1330             d2               = _mm_mul_ps(d,d);
1331             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)))))));
1332
1333             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1334
1335             /* Evaluate switch function */
1336             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1337             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1338             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1339             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1340
1341             fscal            = _mm_add_ps(felec,fvdw);
1342
1343             fscal            = _mm_and_ps(fscal,cutoff_mask);
1344
1345             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1346
1347             /* Calculate temporary vectorial force */
1348             tx               = _mm_mul_ps(fscal,dx00);
1349             ty               = _mm_mul_ps(fscal,dy00);
1350             tz               = _mm_mul_ps(fscal,dz00);
1351
1352             /* Update vectorial force */
1353             fix0             = _mm_add_ps(fix0,tx);
1354             fiy0             = _mm_add_ps(fiy0,ty);
1355             fiz0             = _mm_add_ps(fiz0,tz);
1356
1357             fjx0             = _mm_add_ps(fjx0,tx);
1358             fjy0             = _mm_add_ps(fjy0,ty);
1359             fjz0             = _mm_add_ps(fjz0,tz);
1360
1361             }
1362
1363             /**************************
1364              * CALCULATE INTERACTIONS *
1365              **************************/
1366
1367             if (gmx_mm_any_lt(rsq10,rcutoff2))
1368             {
1369
1370             r10              = _mm_mul_ps(rsq10,rinv10);
1371             r10              = _mm_andnot_ps(dummy_mask,r10);
1372
1373             /* Compute parameters for interactions between i and j atoms */
1374             qq10             = _mm_mul_ps(iq1,jq0);
1375
1376             /* EWALD ELECTROSTATICS */
1377
1378             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1379             ewrt             = _mm_mul_ps(r10,ewtabscale);
1380             ewitab           = _mm_cvttps_epi32(ewrt);
1381             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1382             ewitab           = _mm_slli_epi32(ewitab,2);
1383             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1384             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1385             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1386             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1387             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1388             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1389             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1390             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1391             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1392
1393             d                = _mm_sub_ps(r10,rswitch);
1394             d                = _mm_max_ps(d,_mm_setzero_ps());
1395             d2               = _mm_mul_ps(d,d);
1396             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)))))));
1397
1398             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1399
1400             /* Evaluate switch function */
1401             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1402             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1403             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1404
1405             fscal            = felec;
1406
1407             fscal            = _mm_and_ps(fscal,cutoff_mask);
1408
1409             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1410
1411             /* Calculate temporary vectorial force */
1412             tx               = _mm_mul_ps(fscal,dx10);
1413             ty               = _mm_mul_ps(fscal,dy10);
1414             tz               = _mm_mul_ps(fscal,dz10);
1415
1416             /* Update vectorial force */
1417             fix1             = _mm_add_ps(fix1,tx);
1418             fiy1             = _mm_add_ps(fiy1,ty);
1419             fiz1             = _mm_add_ps(fiz1,tz);
1420
1421             fjx0             = _mm_add_ps(fjx0,tx);
1422             fjy0             = _mm_add_ps(fjy0,ty);
1423             fjz0             = _mm_add_ps(fjz0,tz);
1424
1425             }
1426
1427             /**************************
1428              * CALCULATE INTERACTIONS *
1429              **************************/
1430
1431             if (gmx_mm_any_lt(rsq20,rcutoff2))
1432             {
1433
1434             r20              = _mm_mul_ps(rsq20,rinv20);
1435             r20              = _mm_andnot_ps(dummy_mask,r20);
1436
1437             /* Compute parameters for interactions between i and j atoms */
1438             qq20             = _mm_mul_ps(iq2,jq0);
1439
1440             /* EWALD ELECTROSTATICS */
1441
1442             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1443             ewrt             = _mm_mul_ps(r20,ewtabscale);
1444             ewitab           = _mm_cvttps_epi32(ewrt);
1445             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1446             ewitab           = _mm_slli_epi32(ewitab,2);
1447             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1448             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1449             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1450             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1451             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1452             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1453             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1454             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1455             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1456
1457             d                = _mm_sub_ps(r20,rswitch);
1458             d                = _mm_max_ps(d,_mm_setzero_ps());
1459             d2               = _mm_mul_ps(d,d);
1460             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)))))));
1461
1462             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1463
1464             /* Evaluate switch function */
1465             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1466             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1467             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1468
1469             fscal            = felec;
1470
1471             fscal            = _mm_and_ps(fscal,cutoff_mask);
1472
1473             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1474
1475             /* Calculate temporary vectorial force */
1476             tx               = _mm_mul_ps(fscal,dx20);
1477             ty               = _mm_mul_ps(fscal,dy20);
1478             tz               = _mm_mul_ps(fscal,dz20);
1479
1480             /* Update vectorial force */
1481             fix2             = _mm_add_ps(fix2,tx);
1482             fiy2             = _mm_add_ps(fiy2,ty);
1483             fiz2             = _mm_add_ps(fiz2,tz);
1484
1485             fjx0             = _mm_add_ps(fjx0,tx);
1486             fjy0             = _mm_add_ps(fjy0,ty);
1487             fjz0             = _mm_add_ps(fjz0,tz);
1488
1489             }
1490
1491             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1492             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1493             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1494             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1495
1496             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1497
1498             /* Inner loop uses 204 flops */
1499         }
1500
1501         /* End of innermost loop */
1502
1503         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1504                                               f+i_coord_offset,fshift+i_shift_offset);
1505
1506         /* Increment number of inner iterations */
1507         inneriter                  += j_index_end - j_index_start;
1508
1509         /* Outer loop uses 18 flops */
1510     }
1511
1512     /* Increment number of outer iterations */
1513     outeriter        += nri;
1514
1515     /* Update outer/inner flops */
1516
1517     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*204);
1518 }