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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_GeomW3P1_VF_sse4_1_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          ewitab;
 103     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     real             *ewtab;
 105     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 106     real             rswitch_scalar,d_scalar;
 107     __m128           dummy_mask,cutoff_mask;
 108     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 109     __m128           one     = _mm_set1_ps(1.0);
 110     __m128           two     = _mm_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 136     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 137     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 143     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 144
 145     rswitch_scalar   = fr->rcoulomb_switch;
 146     rswitch          = _mm_set1_ps(rswitch_scalar);
 147     /* Setup switch parameters */
 148     d_scalar         = rcutoff_scalar-rswitch_scalar;
 149     d                = _mm_set1_ps(d_scalar);
 150     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 151     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 152     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 153     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 154     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 155     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 156
 157     /* Avoid stupid compiler warnings */
 158     jnrA = jnrB = jnrC = jnrD = 0;
 159     j_coord_offsetA = 0;
 160     j_coord_offsetB = 0;
 161     j_coord_offsetC = 0;
 162     j_coord_offsetD = 0;
 163
 164     outeriter        = 0;
 165     inneriter        = 0;
 166
 167     for(iidx=0;iidx<4*DIM;iidx++)
 168     {
 169         scratch[iidx] = 0.0;
 170     }
 171
 172     /* Start outer loop over neighborlists */
 173     for(iidx=0; iidx<nri; iidx++)
 174     {
 175         /* Load shift vector for this list */
 176         i_shift_offset   = DIM*shiftidx[iidx];
 177
 178         /* Load limits for loop over neighbors */
 179         j_index_start    = jindex[iidx];
 180         j_index_end      = jindex[iidx+1];
 181
 182         /* Get outer coordinate index */
 183         inr              = iinr[iidx];
 184         i_coord_offset   = DIM*inr;
 185
 186         /* Load i particle coords and add shift vector */
 187         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 188                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 189
 190         fix0             = _mm_setzero_ps();
 191         fiy0             = _mm_setzero_ps();
 192         fiz0             = _mm_setzero_ps();
 193         fix1             = _mm_setzero_ps();
 194         fiy1             = _mm_setzero_ps();
 195         fiz1             = _mm_setzero_ps();
 196         fix2             = _mm_setzero_ps();
 197         fiy2             = _mm_setzero_ps();
 198         fiz2             = _mm_setzero_ps();
 199
 200         /* Reset potential sums */
 201         velecsum         = _mm_setzero_ps();
 202         vvdwsum          = _mm_setzero_ps();
 203
 204         /* Start inner kernel loop */
 205         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 206         {
 207
 208             /* Get j neighbor index, and coordinate index */
 209             jnrA             = jjnr[jidx];
 210             jnrB             = jjnr[jidx+1];
 211             jnrC             = jjnr[jidx+2];
 212             jnrD             = jjnr[jidx+3];
 213             j_coord_offsetA  = DIM*jnrA;
 214             j_coord_offsetB  = DIM*jnrB;
 215             j_coord_offsetC  = DIM*jnrC;
 216             j_coord_offsetD  = DIM*jnrD;
 217
 218             /* load j atom coordinates */
 219             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 220                                               x+j_coord_offsetC,x+j_coord_offsetD,
 221                                               &jx0,&jy0,&jz0);
 222
 223             /* Calculate displacement vector */
 224             dx00             = _mm_sub_ps(ix0,jx0);
 225             dy00             = _mm_sub_ps(iy0,jy0);
 226             dz00             = _mm_sub_ps(iz0,jz0);
 227             dx10             = _mm_sub_ps(ix1,jx0);
 228             dy10             = _mm_sub_ps(iy1,jy0);
 229             dz10             = _mm_sub_ps(iz1,jz0);
 230             dx20             = _mm_sub_ps(ix2,jx0);
 231             dy20             = _mm_sub_ps(iy2,jy0);
 232             dz20             = _mm_sub_ps(iz2,jz0);
 233
 234             /* Calculate squared distance and things based on it */
 235             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 236             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 237             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 238
 239             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 240             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 241             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 242
 243             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 244             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 245             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 246
 247             /* Load parameters for j particles */
 248             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 249                                                               charge+jnrC+0,charge+jnrD+0);
 250             vdwjidx0A        = 2*vdwtype[jnrA+0];
 251             vdwjidx0B        = 2*vdwtype[jnrB+0];
 252             vdwjidx0C        = 2*vdwtype[jnrC+0];
 253             vdwjidx0D        = 2*vdwtype[jnrD+0];
 254
 255             fjx0             = _mm_setzero_ps();
 256             fjy0             = _mm_setzero_ps();
 257             fjz0             = _mm_setzero_ps();
 258
 259             /**************************
 260              * CALCULATE INTERACTIONS *
 261              **************************/
 262
 263             if (gmx_mm_any_lt(rsq00,rcutoff2))
 264             {
 265
 266             r00              = _mm_mul_ps(rsq00,rinv00);
 267
 268             /* Compute parameters for interactions between i and j atoms */
 269             qq00             = _mm_mul_ps(iq0,jq0);
 270             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 271                                          vdwparam+vdwioffset0+vdwjidx0B,
 272                                          vdwparam+vdwioffset0+vdwjidx0C,
 273                                          vdwparam+vdwioffset0+vdwjidx0D,
 274                                          &c6_00,&c12_00);
 275
 276             /* EWALD ELECTROSTATICS */
 277
 278             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 279             ewrt             = _mm_mul_ps(r00,ewtabscale);
 280             ewitab           = _mm_cvttps_epi32(ewrt);
 281             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 282             ewitab           = _mm_slli_epi32(ewitab,2);
 283             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 284             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 285             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 286             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 287             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 288             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 289             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 290             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 291             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 292
 293             /* LENNARD-JONES DISPERSION/REPULSION */
 294
 295             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 296             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 297             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 298             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 299             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 300
 301             d                = _mm_sub_ps(r00,rswitch);
 302             d                = _mm_max_ps(d,_mm_setzero_ps());
 303             d2               = _mm_mul_ps(d,d);
 304             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)))))));
 305
 306             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 307
 308             /* Evaluate switch function */
 309             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 310             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 311             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 312             velec            = _mm_mul_ps(velec,sw);
 313             vvdw             = _mm_mul_ps(vvdw,sw);
 314             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velec            = _mm_and_ps(velec,cutoff_mask);
 318             velecsum         = _mm_add_ps(velecsum,velec);
 319             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 320             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 321
 322             fscal            = _mm_add_ps(felec,fvdw);
 323
 324             fscal            = _mm_and_ps(fscal,cutoff_mask);
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm_mul_ps(fscal,dx00);
 328             ty               = _mm_mul_ps(fscal,dy00);
 329             tz               = _mm_mul_ps(fscal,dz00);
 330
 331             /* Update vectorial force */
 332             fix0             = _mm_add_ps(fix0,tx);
 333             fiy0             = _mm_add_ps(fiy0,ty);
 334             fiz0             = _mm_add_ps(fiz0,tz);
 335
 336             fjx0             = _mm_add_ps(fjx0,tx);
 337             fjy0             = _mm_add_ps(fjy0,ty);
 338             fjz0             = _mm_add_ps(fjz0,tz);
 339
 340             }
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             if (gmx_mm_any_lt(rsq10,rcutoff2))
 347             {
 348
 349             r10              = _mm_mul_ps(rsq10,rinv10);
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq10             = _mm_mul_ps(iq1,jq0);
 353
 354             /* EWALD ELECTROSTATICS */
 355
 356             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 357             ewrt             = _mm_mul_ps(r10,ewtabscale);
 358             ewitab           = _mm_cvttps_epi32(ewrt);
 359             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 360             ewitab           = _mm_slli_epi32(ewitab,2);
 361             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 362             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 363             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 364             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 365             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 366             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 367             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 368             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 369             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 370
 371             d                = _mm_sub_ps(r10,rswitch);
 372             d                = _mm_max_ps(d,_mm_setzero_ps());
 373             d2               = _mm_mul_ps(d,d);
 374             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)))))));
 375
 376             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 377
 378             /* Evaluate switch function */
 379             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 380             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 381             velec            = _mm_mul_ps(velec,sw);
 382             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velec            = _mm_and_ps(velec,cutoff_mask);
 386             velecsum         = _mm_add_ps(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390             fscal            = _mm_and_ps(fscal,cutoff_mask);
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm_mul_ps(fscal,dx10);
 394             ty               = _mm_mul_ps(fscal,dy10);
 395             tz               = _mm_mul_ps(fscal,dz10);
 396
 397             /* Update vectorial force */
 398             fix1             = _mm_add_ps(fix1,tx);
 399             fiy1             = _mm_add_ps(fiy1,ty);
 400             fiz1             = _mm_add_ps(fiz1,tz);
 401
 402             fjx0             = _mm_add_ps(fjx0,tx);
 403             fjy0             = _mm_add_ps(fjy0,ty);
 404             fjz0             = _mm_add_ps(fjz0,tz);
 405
 406             }
 407
 408             /**************************
 409              * CALCULATE INTERACTIONS *
 410              **************************/
 411
 412             if (gmx_mm_any_lt(rsq20,rcutoff2))
 413             {
 414
 415             r20              = _mm_mul_ps(rsq20,rinv20);
 416
 417             /* Compute parameters for interactions between i and j atoms */
 418             qq20             = _mm_mul_ps(iq2,jq0);
 419
 420             /* EWALD ELECTROSTATICS */
 421
 422             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 423             ewrt             = _mm_mul_ps(r20,ewtabscale);
 424             ewitab           = _mm_cvttps_epi32(ewrt);
 425             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 426             ewitab           = _mm_slli_epi32(ewitab,2);
 427             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 428             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 429             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 430             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 431             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 432             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 433             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 434             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 435             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 436
 437             d                = _mm_sub_ps(r20,rswitch);
 438             d                = _mm_max_ps(d,_mm_setzero_ps());
 439             d2               = _mm_mul_ps(d,d);
 440             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)))))));
 441
 442             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 443
 444             /* Evaluate switch function */
 445             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 446             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 447             velec            = _mm_mul_ps(velec,sw);
 448             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             velec            = _mm_and_ps(velec,cutoff_mask);
 452             velecsum         = _mm_add_ps(velecsum,velec);
 453
 454             fscal            = felec;
 455
 456             fscal            = _mm_and_ps(fscal,cutoff_mask);
 457
 458             /* Calculate temporary vectorial force */
 459             tx               = _mm_mul_ps(fscal,dx20);
 460             ty               = _mm_mul_ps(fscal,dy20);
 461             tz               = _mm_mul_ps(fscal,dz20);
 462
 463             /* Update vectorial force */
 464             fix2             = _mm_add_ps(fix2,tx);
 465             fiy2             = _mm_add_ps(fiy2,ty);
 466             fiz2             = _mm_add_ps(fiz2,tz);
 467
 468             fjx0             = _mm_add_ps(fjx0,tx);
 469             fjy0             = _mm_add_ps(fjy0,ty);
 470             fjz0             = _mm_add_ps(fjz0,tz);
 471
 472             }
 473
 474             fjptrA             = f+j_coord_offsetA;
 475             fjptrB             = f+j_coord_offsetB;
 476             fjptrC             = f+j_coord_offsetC;
 477             fjptrD             = f+j_coord_offsetD;
 478
 479             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 480
 481             /* Inner loop uses 213 flops */
 482         }
 483
 484         if(jidx<j_index_end)
 485         {
 486
 487             /* Get j neighbor index, and coordinate index */
 488             jnrlistA         = jjnr[jidx];
 489             jnrlistB         = jjnr[jidx+1];
 490             jnrlistC         = jjnr[jidx+2];
 491             jnrlistD         = jjnr[jidx+3];
 492             /* Sign of each element will be negative for non-real atoms.
 493              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 494              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 495              */
 496             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 497             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 498             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 499             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 500             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 501             j_coord_offsetA  = DIM*jnrA;
 502             j_coord_offsetB  = DIM*jnrB;
 503             j_coord_offsetC  = DIM*jnrC;
 504             j_coord_offsetD  = DIM*jnrD;
 505
 506             /* load j atom coordinates */
 507             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 508                                               x+j_coord_offsetC,x+j_coord_offsetD,
 509                                               &jx0,&jy0,&jz0);
 510
 511             /* Calculate displacement vector */
 512             dx00             = _mm_sub_ps(ix0,jx0);
 513             dy00             = _mm_sub_ps(iy0,jy0);
 514             dz00             = _mm_sub_ps(iz0,jz0);
 515             dx10             = _mm_sub_ps(ix1,jx0);
 516             dy10             = _mm_sub_ps(iy1,jy0);
 517             dz10             = _mm_sub_ps(iz1,jz0);
 518             dx20             = _mm_sub_ps(ix2,jx0);
 519             dy20             = _mm_sub_ps(iy2,jy0);
 520             dz20             = _mm_sub_ps(iz2,jz0);
 521
 522             /* Calculate squared distance and things based on it */
 523             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 524             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 525             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 526
 527             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 528             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 529             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 530
 531             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 532             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 533             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 534
 535             /* Load parameters for j particles */
 536             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 537                                                               charge+jnrC+0,charge+jnrD+0);
 538             vdwjidx0A        = 2*vdwtype[jnrA+0];
 539             vdwjidx0B        = 2*vdwtype[jnrB+0];
 540             vdwjidx0C        = 2*vdwtype[jnrC+0];
 541             vdwjidx0D        = 2*vdwtype[jnrD+0];
 542
 543             fjx0             = _mm_setzero_ps();
 544             fjy0             = _mm_setzero_ps();
 545             fjz0             = _mm_setzero_ps();
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             if (gmx_mm_any_lt(rsq00,rcutoff2))
 552             {
 553
 554             r00              = _mm_mul_ps(rsq00,rinv00);
 555             r00              = _mm_andnot_ps(dummy_mask,r00);
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             qq00             = _mm_mul_ps(iq0,jq0);
 559             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 560                                          vdwparam+vdwioffset0+vdwjidx0B,
 561                                          vdwparam+vdwioffset0+vdwjidx0C,
 562                                          vdwparam+vdwioffset0+vdwjidx0D,
 563                                          &c6_00,&c12_00);
 564
 565             /* EWALD ELECTROSTATICS */
 566
 567             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 568             ewrt             = _mm_mul_ps(r00,ewtabscale);
 569             ewitab           = _mm_cvttps_epi32(ewrt);
 570             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 571             ewitab           = _mm_slli_epi32(ewitab,2);
 572             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 573             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 574             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 575             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 576             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 577             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 578             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 579             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 580             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 581
 582             /* LENNARD-JONES DISPERSION/REPULSION */
 583
 584             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 585             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 586             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 587             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 588             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 589
 590             d                = _mm_sub_ps(r00,rswitch);
 591             d                = _mm_max_ps(d,_mm_setzero_ps());
 592             d2               = _mm_mul_ps(d,d);
 593             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)))))));
 594
 595             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 596
 597             /* Evaluate switch function */
 598             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 599             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 600             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 601             velec            = _mm_mul_ps(velec,sw);
 602             vvdw             = _mm_mul_ps(vvdw,sw);
 603             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm_and_ps(velec,cutoff_mask);
 607             velec            = _mm_andnot_ps(dummy_mask,velec);
 608             velecsum         = _mm_add_ps(velecsum,velec);
 609             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 610             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 611             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 612
 613             fscal            = _mm_add_ps(felec,fvdw);
 614
 615             fscal            = _mm_and_ps(fscal,cutoff_mask);
 616
 617             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm_mul_ps(fscal,dx00);
 621             ty               = _mm_mul_ps(fscal,dy00);
 622             tz               = _mm_mul_ps(fscal,dz00);
 623
 624             /* Update vectorial force */
 625             fix0             = _mm_add_ps(fix0,tx);
 626             fiy0             = _mm_add_ps(fiy0,ty);
 627             fiz0             = _mm_add_ps(fiz0,tz);
 628
 629             fjx0             = _mm_add_ps(fjx0,tx);
 630             fjy0             = _mm_add_ps(fjy0,ty);
 631             fjz0             = _mm_add_ps(fjz0,tz);
 632
 633             }
 634
 635             /**************************
 636              * CALCULATE INTERACTIONS *
 637              **************************/
 638
 639             if (gmx_mm_any_lt(rsq10,rcutoff2))
 640             {
 641
 642             r10              = _mm_mul_ps(rsq10,rinv10);
 643             r10              = _mm_andnot_ps(dummy_mask,r10);
 644
 645             /* Compute parameters for interactions between i and j atoms */
 646             qq10             = _mm_mul_ps(iq1,jq0);
 647
 648             /* EWALD ELECTROSTATICS */
 649
 650             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 651             ewrt             = _mm_mul_ps(r10,ewtabscale);
 652             ewitab           = _mm_cvttps_epi32(ewrt);
 653             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 654             ewitab           = _mm_slli_epi32(ewitab,2);
 655             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 656             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 657             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 658             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 659             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 660             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 661             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 662             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 663             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 664
 665             d                = _mm_sub_ps(r10,rswitch);
 666             d                = _mm_max_ps(d,_mm_setzero_ps());
 667             d2               = _mm_mul_ps(d,d);
 668             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)))))));
 669
 670             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 671
 672             /* Evaluate switch function */
 673             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 674             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 675             velec            = _mm_mul_ps(velec,sw);
 676             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 677
 678             /* Update potential sum for this i atom from the interaction with this j atom. */
 679             velec            = _mm_and_ps(velec,cutoff_mask);
 680             velec            = _mm_andnot_ps(dummy_mask,velec);
 681             velecsum         = _mm_add_ps(velecsum,velec);
 682
 683             fscal            = felec;
 684
 685             fscal            = _mm_and_ps(fscal,cutoff_mask);
 686
 687             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 688
 689             /* Calculate temporary vectorial force */
 690             tx               = _mm_mul_ps(fscal,dx10);
 691             ty               = _mm_mul_ps(fscal,dy10);
 692             tz               = _mm_mul_ps(fscal,dz10);
 693
 694             /* Update vectorial force */
 695             fix1             = _mm_add_ps(fix1,tx);
 696             fiy1             = _mm_add_ps(fiy1,ty);
 697             fiz1             = _mm_add_ps(fiz1,tz);
 698
 699             fjx0             = _mm_add_ps(fjx0,tx);
 700             fjy0             = _mm_add_ps(fjy0,ty);
 701             fjz0             = _mm_add_ps(fjz0,tz);
 702
 703             }
 704
 705             /**************************
 706              * CALCULATE INTERACTIONS *
 707              **************************/
 708
 709             if (gmx_mm_any_lt(rsq20,rcutoff2))
 710             {
 711
 712             r20              = _mm_mul_ps(rsq20,rinv20);
 713             r20              = _mm_andnot_ps(dummy_mask,r20);
 714
 715             /* Compute parameters for interactions between i and j atoms */
 716             qq20             = _mm_mul_ps(iq2,jq0);
 717
 718             /* EWALD ELECTROSTATICS */
 719
 720             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 721             ewrt             = _mm_mul_ps(r20,ewtabscale);
 722             ewitab           = _mm_cvttps_epi32(ewrt);
 723             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 724             ewitab           = _mm_slli_epi32(ewitab,2);
 725             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 726             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 727             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 728             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 729             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 730             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 731             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 732             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 733             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 734
 735             d                = _mm_sub_ps(r20,rswitch);
 736             d                = _mm_max_ps(d,_mm_setzero_ps());
 737             d2               = _mm_mul_ps(d,d);
 738             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)))))));
 739
 740             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 741
 742             /* Evaluate switch function */
 743             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 744             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 745             velec            = _mm_mul_ps(velec,sw);
 746             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 747
 748             /* Update potential sum for this i atom from the interaction with this j atom. */
 749             velec            = _mm_and_ps(velec,cutoff_mask);
 750             velec            = _mm_andnot_ps(dummy_mask,velec);
 751             velecsum         = _mm_add_ps(velecsum,velec);
 752
 753             fscal            = felec;
 754
 755             fscal            = _mm_and_ps(fscal,cutoff_mask);
 756
 757             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 758
 759             /* Calculate temporary vectorial force */
 760             tx               = _mm_mul_ps(fscal,dx20);
 761             ty               = _mm_mul_ps(fscal,dy20);
 762             tz               = _mm_mul_ps(fscal,dz20);
 763
 764             /* Update vectorial force */
 765             fix2             = _mm_add_ps(fix2,tx);
 766             fiy2             = _mm_add_ps(fiy2,ty);
 767             fiz2             = _mm_add_ps(fiz2,tz);
 768
 769             fjx0             = _mm_add_ps(fjx0,tx);
 770             fjy0             = _mm_add_ps(fjy0,ty);
 771             fjz0             = _mm_add_ps(fjz0,tz);
 772
 773             }
 774
 775             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 776             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 777             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 778             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 779
 780             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 781
 782             /* Inner loop uses 216 flops */
 783         }
 784
 785         /* End of innermost loop */
 786
 787         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 788                                               f+i_coord_offset,fshift+i_shift_offset);
 789
 790         ggid                        = gid[iidx];
 791         /* Update potential energies */
 792         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 793         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 794
 795         /* Increment number of inner iterations */
 796         inneriter                  += j_index_end - j_index_start;
 797
 798         /* Outer loop uses 20 flops */
 799     }
 800
 801     /* Increment number of outer iterations */
 802     outeriter        += nri;
 803
 804     /* Update outer/inner flops */
 805
 806     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*216);
 807 }
 808 /*
 809  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_sse4_1_single
 810  * Electrostatics interaction: Ewald
 811  * VdW interaction:            LennardJones
 812  * Geometry:                   Water3-Particle
 813  * Calculate force/pot:        Force
 814  */
 815 void
 816 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_sse4_1_single
 817                     (t_nblist                    * gmx_restrict       nlist,
 818                      rvec                        * gmx_restrict          xx,
 819                      rvec                        * gmx_restrict          ff,
 820                      t_forcerec                  * gmx_restrict          fr,
 821                      t_mdatoms                   * gmx_restrict     mdatoms,
 822                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 823                      t_nrnb                      * gmx_restrict        nrnb)
 824 {
 825     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 826      * just 0 for non-waters.
 827      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 828      * jnr indices corresponding to data put in the four positions in the SIMD register.
 829      */
 830     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 831     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 832     int              jnrA,jnrB,jnrC,jnrD;
 833     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 834     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 835     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 836     real             rcutoff_scalar;
 837     real             *shiftvec,*fshift,*x,*f;
 838     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 839     real             scratch[4*DIM];
 840     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 841     int              vdwioffset0;
 842     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 843     int              vdwioffset1;
 844     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 845     int              vdwioffset2;
 846     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 847     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 848     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 849     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 850     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 851     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 852     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 853     real             *charge;
 854     int              nvdwtype;
 855     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 856     int              *vdwtype;
 857     real             *vdwparam;
 858     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 859     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 860     __m128i          ewitab;
 861     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 862     real             *ewtab;
 863     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 864     real             rswitch_scalar,d_scalar;
 865     __m128           dummy_mask,cutoff_mask;
 866     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 867     __m128           one     = _mm_set1_ps(1.0);
 868     __m128           two     = _mm_set1_ps(2.0);
 869     x                = xx[0];
 870     f                = ff[0];
 871
 872     nri              = nlist->nri;
 873     iinr             = nlist->iinr;
 874     jindex           = nlist->jindex;
 875     jjnr             = nlist->jjnr;
 876     shiftidx         = nlist->shift;
 877     gid              = nlist->gid;
 878     shiftvec         = fr->shift_vec[0];
 879     fshift           = fr->fshift[0];
 880     facel            = _mm_set1_ps(fr->epsfac);
 881     charge           = mdatoms->chargeA;
 882     nvdwtype         = fr->ntype;
 883     vdwparam         = fr->nbfp;
 884     vdwtype          = mdatoms->typeA;
 885
 886     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 887     ewtab            = fr->ic->tabq_coul_FDV0;
 888     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 889     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 890
 891     /* Setup water-specific parameters */
 892     inr              = nlist->iinr[0];
 893     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 894     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 895     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 896     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 897
 898     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 899     rcutoff_scalar   = fr->rcoulomb;
 900     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 901     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 902
 903     rswitch_scalar   = fr->rcoulomb_switch;
 904     rswitch          = _mm_set1_ps(rswitch_scalar);
 905     /* Setup switch parameters */
 906     d_scalar         = rcutoff_scalar-rswitch_scalar;
 907     d                = _mm_set1_ps(d_scalar);
 908     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 909     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 910     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 911     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 912     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 913     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 914
 915     /* Avoid stupid compiler warnings */
 916     jnrA = jnrB = jnrC = jnrD = 0;
 917     j_coord_offsetA = 0;
 918     j_coord_offsetB = 0;
 919     j_coord_offsetC = 0;
 920     j_coord_offsetD = 0;
 921
 922     outeriter        = 0;
 923     inneriter        = 0;
 924
 925     for(iidx=0;iidx<4*DIM;iidx++)
 926     {
 927         scratch[iidx] = 0.0;
 928     }
 929
 930     /* Start outer loop over neighborlists */
 931     for(iidx=0; iidx<nri; iidx++)
 932     {
 933         /* Load shift vector for this list */
 934         i_shift_offset   = DIM*shiftidx[iidx];
 935
 936         /* Load limits for loop over neighbors */
 937         j_index_start    = jindex[iidx];
 938         j_index_end      = jindex[iidx+1];
 939
 940         /* Get outer coordinate index */
 941         inr              = iinr[iidx];
 942         i_coord_offset   = DIM*inr;
 943
 944         /* Load i particle coords and add shift vector */
 945         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 946                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 947
 948         fix0             = _mm_setzero_ps();
 949         fiy0             = _mm_setzero_ps();
 950         fiz0             = _mm_setzero_ps();
 951         fix1             = _mm_setzero_ps();
 952         fiy1             = _mm_setzero_ps();
 953         fiz1             = _mm_setzero_ps();
 954         fix2             = _mm_setzero_ps();
 955         fiy2             = _mm_setzero_ps();
 956         fiz2             = _mm_setzero_ps();
 957
 958         /* Start inner kernel loop */
 959         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 960         {
 961
 962             /* Get j neighbor index, and coordinate index */
 963             jnrA             = jjnr[jidx];
 964             jnrB             = jjnr[jidx+1];
 965             jnrC             = jjnr[jidx+2];
 966             jnrD             = jjnr[jidx+3];
 967             j_coord_offsetA  = DIM*jnrA;
 968             j_coord_offsetB  = DIM*jnrB;
 969             j_coord_offsetC  = DIM*jnrC;
 970             j_coord_offsetD  = DIM*jnrD;
 971
 972             /* load j atom coordinates */
 973             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 974                                               x+j_coord_offsetC,x+j_coord_offsetD,
 975                                               &jx0,&jy0,&jz0);
 976
 977             /* Calculate displacement vector */
 978             dx00             = _mm_sub_ps(ix0,jx0);
 979             dy00             = _mm_sub_ps(iy0,jy0);
 980             dz00             = _mm_sub_ps(iz0,jz0);
 981             dx10             = _mm_sub_ps(ix1,jx0);
 982             dy10             = _mm_sub_ps(iy1,jy0);
 983             dz10             = _mm_sub_ps(iz1,jz0);
 984             dx20             = _mm_sub_ps(ix2,jx0);
 985             dy20             = _mm_sub_ps(iy2,jy0);
 986             dz20             = _mm_sub_ps(iz2,jz0);
 987
 988             /* Calculate squared distance and things based on it */
 989             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 990             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 991             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 992
 993             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 994             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 995             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 996
 997             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 998             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 999             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1000
1001             /* Load parameters for j particles */
1002             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1003                                                               charge+jnrC+0,charge+jnrD+0);
1004             vdwjidx0A        = 2*vdwtype[jnrA+0];
1005             vdwjidx0B        = 2*vdwtype[jnrB+0];
1006             vdwjidx0C        = 2*vdwtype[jnrC+0];
1007             vdwjidx0D        = 2*vdwtype[jnrD+0];
1008
1009             fjx0             = _mm_setzero_ps();
1010             fjy0             = _mm_setzero_ps();
1011             fjz0             = _mm_setzero_ps();
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             if (gmx_mm_any_lt(rsq00,rcutoff2))
1018             {
1019
1020             r00              = _mm_mul_ps(rsq00,rinv00);
1021
1022             /* Compute parameters for interactions between i and j atoms */
1023             qq00             = _mm_mul_ps(iq0,jq0);
1024             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1025                                          vdwparam+vdwioffset0+vdwjidx0B,
1026                                          vdwparam+vdwioffset0+vdwjidx0C,
1027                                          vdwparam+vdwioffset0+vdwjidx0D,
1028                                          &c6_00,&c12_00);
1029
1030             /* EWALD ELECTROSTATICS */
1031
1032             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1033             ewrt             = _mm_mul_ps(r00,ewtabscale);
1034             ewitab           = _mm_cvttps_epi32(ewrt);
1035             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1036             ewitab           = _mm_slli_epi32(ewitab,2);
1037             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1038             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1039             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1040             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1041             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1042             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1043             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1044             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
1045             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1046
1047             /* LENNARD-JONES DISPERSION/REPULSION */
1048
1049             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1050             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1051             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1052             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1053             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1054
1055             d                = _mm_sub_ps(r00,rswitch);
1056             d                = _mm_max_ps(d,_mm_setzero_ps());
1057             d2               = _mm_mul_ps(d,d);
1058             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)))))));
1059
1060             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1061
1062             /* Evaluate switch function */
1063             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1064             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1065             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1066             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1067
1068             fscal            = _mm_add_ps(felec,fvdw);
1069
1070             fscal            = _mm_and_ps(fscal,cutoff_mask);
1071
1072             /* Calculate temporary vectorial force */
1073             tx               = _mm_mul_ps(fscal,dx00);
1074             ty               = _mm_mul_ps(fscal,dy00);
1075             tz               = _mm_mul_ps(fscal,dz00);
1076
1077             /* Update vectorial force */
1078             fix0             = _mm_add_ps(fix0,tx);
1079             fiy0             = _mm_add_ps(fiy0,ty);
1080             fiz0             = _mm_add_ps(fiz0,tz);
1081
1082             fjx0             = _mm_add_ps(fjx0,tx);
1083             fjy0             = _mm_add_ps(fjy0,ty);
1084             fjz0             = _mm_add_ps(fjz0,tz);
1085
1086             }
1087
1088             /**************************
1089              * CALCULATE INTERACTIONS *
1090              **************************/
1091
1092             if (gmx_mm_any_lt(rsq10,rcutoff2))
1093             {
1094
1095             r10              = _mm_mul_ps(rsq10,rinv10);
1096
1097             /* Compute parameters for interactions between i and j atoms */
1098             qq10             = _mm_mul_ps(iq1,jq0);
1099
1100             /* EWALD ELECTROSTATICS */
1101
1102             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1103             ewrt             = _mm_mul_ps(r10,ewtabscale);
1104             ewitab           = _mm_cvttps_epi32(ewrt);
1105             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1106             ewitab           = _mm_slli_epi32(ewitab,2);
1107             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1108             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1109             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1110             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1111             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1112             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1113             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1114             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1115             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1116
1117             d                = _mm_sub_ps(r10,rswitch);
1118             d                = _mm_max_ps(d,_mm_setzero_ps());
1119             d2               = _mm_mul_ps(d,d);
1120             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)))))));
1121
1122             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1123
1124             /* Evaluate switch function */
1125             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1126             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1127             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1128
1129             fscal            = felec;
1130
1131             fscal            = _mm_and_ps(fscal,cutoff_mask);
1132
1133             /* Calculate temporary vectorial force */
1134             tx               = _mm_mul_ps(fscal,dx10);
1135             ty               = _mm_mul_ps(fscal,dy10);
1136             tz               = _mm_mul_ps(fscal,dz10);
1137
1138             /* Update vectorial force */
1139             fix1             = _mm_add_ps(fix1,tx);
1140             fiy1             = _mm_add_ps(fiy1,ty);
1141             fiz1             = _mm_add_ps(fiz1,tz);
1142
1143             fjx0             = _mm_add_ps(fjx0,tx);
1144             fjy0             = _mm_add_ps(fjy0,ty);
1145             fjz0             = _mm_add_ps(fjz0,tz);
1146
1147             }
1148
1149             /**************************
1150              * CALCULATE INTERACTIONS *
1151              **************************/
1152
1153             if (gmx_mm_any_lt(rsq20,rcutoff2))
1154             {
1155
1156             r20              = _mm_mul_ps(rsq20,rinv20);
1157
1158             /* Compute parameters for interactions between i and j atoms */
1159             qq20             = _mm_mul_ps(iq2,jq0);
1160
1161             /* EWALD ELECTROSTATICS */
1162
1163             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1164             ewrt             = _mm_mul_ps(r20,ewtabscale);
1165             ewitab           = _mm_cvttps_epi32(ewrt);
1166             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1167             ewitab           = _mm_slli_epi32(ewitab,2);
1168             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1169             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1170             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1171             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1172             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1173             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1174             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1175             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1176             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1177
1178             d                = _mm_sub_ps(r20,rswitch);
1179             d                = _mm_max_ps(d,_mm_setzero_ps());
1180             d2               = _mm_mul_ps(d,d);
1181             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)))))));
1182
1183             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1184
1185             /* Evaluate switch function */
1186             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1187             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1188             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1189
1190             fscal            = felec;
1191
1192             fscal            = _mm_and_ps(fscal,cutoff_mask);
1193
1194             /* Calculate temporary vectorial force */
1195             tx               = _mm_mul_ps(fscal,dx20);
1196             ty               = _mm_mul_ps(fscal,dy20);
1197             tz               = _mm_mul_ps(fscal,dz20);
1198
1199             /* Update vectorial force */
1200             fix2             = _mm_add_ps(fix2,tx);
1201             fiy2             = _mm_add_ps(fiy2,ty);
1202             fiz2             = _mm_add_ps(fiz2,tz);
1203
1204             fjx0             = _mm_add_ps(fjx0,tx);
1205             fjy0             = _mm_add_ps(fjy0,ty);
1206             fjz0             = _mm_add_ps(fjz0,tz);
1207
1208             }
1209
1210             fjptrA             = f+j_coord_offsetA;
1211             fjptrB             = f+j_coord_offsetB;
1212             fjptrC             = f+j_coord_offsetC;
1213             fjptrD             = f+j_coord_offsetD;
1214
1215             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1216
1217             /* Inner loop uses 201 flops */
1218         }
1219
1220         if(jidx<j_index_end)
1221         {
1222
1223             /* Get j neighbor index, and coordinate index */
1224             jnrlistA         = jjnr[jidx];
1225             jnrlistB         = jjnr[jidx+1];
1226             jnrlistC         = jjnr[jidx+2];
1227             jnrlistD         = jjnr[jidx+3];
1228             /* Sign of each element will be negative for non-real atoms.
1229              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1230              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1231              */
1232             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1233             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1234             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1235             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1236             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1237             j_coord_offsetA  = DIM*jnrA;
1238             j_coord_offsetB  = DIM*jnrB;
1239             j_coord_offsetC  = DIM*jnrC;
1240             j_coord_offsetD  = DIM*jnrD;
1241
1242             /* load j atom coordinates */
1243             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1244                                               x+j_coord_offsetC,x+j_coord_offsetD,
1245                                               &jx0,&jy0,&jz0);
1246
1247             /* Calculate displacement vector */
1248             dx00             = _mm_sub_ps(ix0,jx0);
1249             dy00             = _mm_sub_ps(iy0,jy0);
1250             dz00             = _mm_sub_ps(iz0,jz0);
1251             dx10             = _mm_sub_ps(ix1,jx0);
1252             dy10             = _mm_sub_ps(iy1,jy0);
1253             dz10             = _mm_sub_ps(iz1,jz0);
1254             dx20             = _mm_sub_ps(ix2,jx0);
1255             dy20             = _mm_sub_ps(iy2,jy0);
1256             dz20             = _mm_sub_ps(iz2,jz0);
1257
1258             /* Calculate squared distance and things based on it */
1259             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1260             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1261             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1262
1263             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1264             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1265             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1266
1267             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1268             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1269             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1270
1271             /* Load parameters for j particles */
1272             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1273                                                               charge+jnrC+0,charge+jnrD+0);
1274             vdwjidx0A        = 2*vdwtype[jnrA+0];
1275             vdwjidx0B        = 2*vdwtype[jnrB+0];
1276             vdwjidx0C        = 2*vdwtype[jnrC+0];
1277             vdwjidx0D        = 2*vdwtype[jnrD+0];
1278
1279             fjx0             = _mm_setzero_ps();
1280             fjy0             = _mm_setzero_ps();
1281             fjz0             = _mm_setzero_ps();
1282
1283             /**************************
1284              * CALCULATE INTERACTIONS *
1285              **************************/
1286
1287             if (gmx_mm_any_lt(rsq00,rcutoff2))
1288             {
1289
1290             r00              = _mm_mul_ps(rsq00,rinv00);
1291             r00              = _mm_andnot_ps(dummy_mask,r00);
1292
1293             /* Compute parameters for interactions between i and j atoms */
1294             qq00             = _mm_mul_ps(iq0,jq0);
1295             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1296                                          vdwparam+vdwioffset0+vdwjidx0B,
1297                                          vdwparam+vdwioffset0+vdwjidx0C,
1298                                          vdwparam+vdwioffset0+vdwjidx0D,
1299                                          &c6_00,&c12_00);
1300
1301             /* EWALD ELECTROSTATICS */
1302
1303             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1304             ewrt             = _mm_mul_ps(r00,ewtabscale);
1305             ewitab           = _mm_cvttps_epi32(ewrt);
1306             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1307             ewitab           = _mm_slli_epi32(ewitab,2);
1308             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1309             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1310             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1311             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1312             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1313             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1314             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1315             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
1316             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1317
1318             /* LENNARD-JONES DISPERSION/REPULSION */
1319
1320             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1321             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1322             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1323             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1324             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1325
1326             d                = _mm_sub_ps(r00,rswitch);
1327             d                = _mm_max_ps(d,_mm_setzero_ps());
1328             d2               = _mm_mul_ps(d,d);
1329             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)))))));
1330
1331             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1332
1333             /* Evaluate switch function */
1334             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1335             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1336             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1337             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1338
1339             fscal            = _mm_add_ps(felec,fvdw);
1340
1341             fscal            = _mm_and_ps(fscal,cutoff_mask);
1342
1343             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1344
1345             /* Calculate temporary vectorial force */
1346             tx               = _mm_mul_ps(fscal,dx00);
1347             ty               = _mm_mul_ps(fscal,dy00);
1348             tz               = _mm_mul_ps(fscal,dz00);
1349
1350             /* Update vectorial force */
1351             fix0             = _mm_add_ps(fix0,tx);
1352             fiy0             = _mm_add_ps(fiy0,ty);
1353             fiz0             = _mm_add_ps(fiz0,tz);
1354
1355             fjx0             = _mm_add_ps(fjx0,tx);
1356             fjy0             = _mm_add_ps(fjy0,ty);
1357             fjz0             = _mm_add_ps(fjz0,tz);
1358
1359             }
1360
1361             /**************************
1362              * CALCULATE INTERACTIONS *
1363              **************************/
1364
1365             if (gmx_mm_any_lt(rsq10,rcutoff2))
1366             {
1367
1368             r10              = _mm_mul_ps(rsq10,rinv10);
1369             r10              = _mm_andnot_ps(dummy_mask,r10);
1370
1371             /* Compute parameters for interactions between i and j atoms */
1372             qq10             = _mm_mul_ps(iq1,jq0);
1373
1374             /* EWALD ELECTROSTATICS */
1375
1376             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1377             ewrt             = _mm_mul_ps(r10,ewtabscale);
1378             ewitab           = _mm_cvttps_epi32(ewrt);
1379             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1380             ewitab           = _mm_slli_epi32(ewitab,2);
1381             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1382             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1383             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1384             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1385             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1386             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1387             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1388             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1389             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1390
1391             d                = _mm_sub_ps(r10,rswitch);
1392             d                = _mm_max_ps(d,_mm_setzero_ps());
1393             d2               = _mm_mul_ps(d,d);
1394             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)))))));
1395
1396             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1397
1398             /* Evaluate switch function */
1399             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1400             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1401             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1402
1403             fscal            = felec;
1404
1405             fscal            = _mm_and_ps(fscal,cutoff_mask);
1406
1407             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1408
1409             /* Calculate temporary vectorial force */
1410             tx               = _mm_mul_ps(fscal,dx10);
1411             ty               = _mm_mul_ps(fscal,dy10);
1412             tz               = _mm_mul_ps(fscal,dz10);
1413
1414             /* Update vectorial force */
1415             fix1             = _mm_add_ps(fix1,tx);
1416             fiy1             = _mm_add_ps(fiy1,ty);
1417             fiz1             = _mm_add_ps(fiz1,tz);
1418
1419             fjx0             = _mm_add_ps(fjx0,tx);
1420             fjy0             = _mm_add_ps(fjy0,ty);
1421             fjz0             = _mm_add_ps(fjz0,tz);
1422
1423             }
1424
1425             /**************************
1426              * CALCULATE INTERACTIONS *
1427              **************************/
1428
1429             if (gmx_mm_any_lt(rsq20,rcutoff2))
1430             {
1431
1432             r20              = _mm_mul_ps(rsq20,rinv20);
1433             r20              = _mm_andnot_ps(dummy_mask,r20);
1434
1435             /* Compute parameters for interactions between i and j atoms */
1436             qq20             = _mm_mul_ps(iq2,jq0);
1437
1438             /* EWALD ELECTROSTATICS */
1439
1440             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1441             ewrt             = _mm_mul_ps(r20,ewtabscale);
1442             ewitab           = _mm_cvttps_epi32(ewrt);
1443             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1444             ewitab           = _mm_slli_epi32(ewitab,2);
1445             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1446             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1447             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1448             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1449             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1450             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1451             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1452             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1453             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1454
1455             d                = _mm_sub_ps(r20,rswitch);
1456             d                = _mm_max_ps(d,_mm_setzero_ps());
1457             d2               = _mm_mul_ps(d,d);
1458             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)))))));
1459
1460             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1461
1462             /* Evaluate switch function */
1463             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1464             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1465             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1466
1467             fscal            = felec;
1468
1469             fscal            = _mm_and_ps(fscal,cutoff_mask);
1470
1471             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1472
1473             /* Calculate temporary vectorial force */
1474             tx               = _mm_mul_ps(fscal,dx20);
1475             ty               = _mm_mul_ps(fscal,dy20);
1476             tz               = _mm_mul_ps(fscal,dz20);
1477
1478             /* Update vectorial force */
1479             fix2             = _mm_add_ps(fix2,tx);
1480             fiy2             = _mm_add_ps(fiy2,ty);
1481             fiz2             = _mm_add_ps(fiz2,tz);
1482
1483             fjx0             = _mm_add_ps(fjx0,tx);
1484             fjy0             = _mm_add_ps(fjy0,ty);
1485             fjz0             = _mm_add_ps(fjz0,tz);
1486
1487             }
1488
1489             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1490             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1491             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1492             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1493
1494             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1495
1496             /* Inner loop uses 204 flops */
1497         }
1498
1499         /* End of innermost loop */
1500
1501         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1502                                               f+i_coord_offset,fshift+i_shift_offset);
1503
1504         /* Increment number of inner iterations */
1505         inneriter                  += j_index_end - j_index_start;
1506
1507         /* Outer loop uses 18 flops */
1508     }
1509
1510     /* Increment number of outer iterations */
1511     outeriter        += nri;
1512
1513     /* Update outer/inner flops */
1514
1515     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*204);
1516 }