src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecGB_VdwCSTab_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 sse2_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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: GeneralizedBorn
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128i          gbitab;
  90     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  91     __m128           minushalf = _mm_set1_ps(-0.5);
  92     real             *invsqrta,*dvda,*gbtab;
  93     int              nvdwtype;
  94     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  98     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m128           dummy_mask,cutoff_mask;
 104     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 105     __m128           one     = _mm_set1_ps(1.0);
 106     __m128           two     = _mm_set1_ps(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_ps(fr->ic->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     vftab            = kernel_data->table_vdw->data;
 125     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 126
 127     invsqrta         = fr->invsqrta;
 128     dvda             = fr->dvda;
 129     gbtabscale       = _mm_set1_ps(fr->gbtab->scale);
 130     gbtab            = fr->gbtab->data;
 131     gbinvepsdiff     = _mm_set1_ps((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     for(iidx=0;iidx<4*DIM;iidx++)
 144     {
 145         scratch[iidx] = 0.0;
 146     }
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 164
 165         fix0             = _mm_setzero_ps();
 166         fiy0             = _mm_setzero_ps();
 167         fiz0             = _mm_setzero_ps();
 168
 169         /* Load parameters for i particles */
 170         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 171         isai0            = _mm_load1_ps(invsqrta+inr+0);
 172         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_ps();
 176         vgbsum           = _mm_setzero_ps();
 177         vvdwsum          = _mm_setzero_ps();
 178         dvdasum          = _mm_setzero_ps();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             jnrC             = jjnr[jidx+2];
 188             jnrD             = jjnr[jidx+3];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               x+j_coord_offsetC,x+j_coord_offsetD,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm_sub_ps(ix0,jx0);
 201             dy00             = _mm_sub_ps(iy0,jy0);
 202             dz00             = _mm_sub_ps(iz0,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 206
 207             rinv00           = sse2_invsqrt_f(rsq00);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 211                                                               charge+jnrC+0,charge+jnrD+0);
 212             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 213                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216             vdwjidx0C        = 2*vdwtype[jnrC+0];
 217             vdwjidx0D        = 2*vdwtype[jnrD+0];
 218
 219             /**************************
 220              * CALCULATE INTERACTIONS *
 221              **************************/
 222
 223             r00              = _mm_mul_ps(rsq00,rinv00);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm_mul_ps(iq0,jq0);
 227             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 228                                          vdwparam+vdwioffset0+vdwjidx0B,
 229                                          vdwparam+vdwioffset0+vdwjidx0C,
 230                                          vdwparam+vdwioffset0+vdwjidx0D,
 231                                          &c6_00,&c12_00);
 232
 233             /* Calculate table index by multiplying r with table scale and truncate to integer */
 234             rt               = _mm_mul_ps(r00,vftabscale);
 235             vfitab           = _mm_cvttps_epi32(rt);
 236             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 237             vfitab           = _mm_slli_epi32(vfitab,3);
 238
 239             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 240             isaprod          = _mm_mul_ps(isai0,isaj0);
 241             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 242             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 243
 244             /* Calculate generalized born table index - this is a separate table from the normal one,
 245              * but we use the same procedure by multiplying r with scale and truncating to integer.
 246              */
 247             rt               = _mm_mul_ps(r00,gbscale);
 248             gbitab           = _mm_cvttps_epi32(rt);
 249             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 250             gbitab           = _mm_slli_epi32(gbitab,2);
 251
 252             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 253             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 254             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 255             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 256             _MM_TRANSPOSE4_PS(Y,F,G,H);
 257             Heps             = _mm_mul_ps(gbeps,H);
 258             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 259             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 260             vgb              = _mm_mul_ps(gbqqfactor,VV);
 261
 262             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 263             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 264             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 265             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 266             fjptrA           = dvda+jnrA;
 267             fjptrB           = dvda+jnrB;
 268             fjptrC           = dvda+jnrC;
 269             fjptrD           = dvda+jnrD;
 270             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 271             velec            = _mm_mul_ps(qq00,rinv00);
 272             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 273
 274             /* CUBIC SPLINE TABLE DISPERSION */
 275             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 276             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 277             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 278             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 279             _MM_TRANSPOSE4_PS(Y,F,G,H);
 280             Heps             = _mm_mul_ps(vfeps,H);
 281             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 282             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 283             vvdw6            = _mm_mul_ps(c6_00,VV);
 284             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 285             fvdw6            = _mm_mul_ps(c6_00,FF);
 286
 287             /* CUBIC SPLINE TABLE REPULSION */
 288             vfitab           = _mm_add_epi32(vfitab,ifour);
 289             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 290             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 291             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 292             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 293             _MM_TRANSPOSE4_PS(Y,F,G,H);
 294             Heps             = _mm_mul_ps(vfeps,H);
 295             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 296             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 297             vvdw12           = _mm_mul_ps(c12_00,VV);
 298             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 299             fvdw12           = _mm_mul_ps(c12_00,FF);
 300             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 301             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velecsum         = _mm_add_ps(velecsum,velec);
 305             vgbsum           = _mm_add_ps(vgbsum,vgb);
 306             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 307
 308             fscal            = _mm_add_ps(felec,fvdw);
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm_mul_ps(fscal,dx00);
 312             ty               = _mm_mul_ps(fscal,dy00);
 313             tz               = _mm_mul_ps(fscal,dz00);
 314
 315             /* Update vectorial force */
 316             fix0             = _mm_add_ps(fix0,tx);
 317             fiy0             = _mm_add_ps(fiy0,ty);
 318             fiz0             = _mm_add_ps(fiz0,tz);
 319
 320             fjptrA             = f+j_coord_offsetA;
 321             fjptrB             = f+j_coord_offsetB;
 322             fjptrC             = f+j_coord_offsetC;
 323             fjptrD             = f+j_coord_offsetD;
 324             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 325
 326             /* Inner loop uses 92 flops */
 327         }
 328
 329         if(jidx<j_index_end)
 330         {
 331
 332             /* Get j neighbor index, and coordinate index */
 333             jnrlistA         = jjnr[jidx];
 334             jnrlistB         = jjnr[jidx+1];
 335             jnrlistC         = jjnr[jidx+2];
 336             jnrlistD         = jjnr[jidx+3];
 337             /* Sign of each element will be negative for non-real atoms.
 338              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 339              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 340              */
 341             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 342             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 343             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 344             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 345             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 346             j_coord_offsetA  = DIM*jnrA;
 347             j_coord_offsetB  = DIM*jnrB;
 348             j_coord_offsetC  = DIM*jnrC;
 349             j_coord_offsetD  = DIM*jnrD;
 350
 351             /* load j atom coordinates */
 352             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 353                                               x+j_coord_offsetC,x+j_coord_offsetD,
 354                                               &jx0,&jy0,&jz0);
 355
 356             /* Calculate displacement vector */
 357             dx00             = _mm_sub_ps(ix0,jx0);
 358             dy00             = _mm_sub_ps(iy0,jy0);
 359             dz00             = _mm_sub_ps(iz0,jz0);
 360
 361             /* Calculate squared distance and things based on it */
 362             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 363
 364             rinv00           = sse2_invsqrt_f(rsq00);
 365
 366             /* Load parameters for j particles */
 367             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 368                                                               charge+jnrC+0,charge+jnrD+0);
 369             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 370                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 371             vdwjidx0A        = 2*vdwtype[jnrA+0];
 372             vdwjidx0B        = 2*vdwtype[jnrB+0];
 373             vdwjidx0C        = 2*vdwtype[jnrC+0];
 374             vdwjidx0D        = 2*vdwtype[jnrD+0];
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             r00              = _mm_mul_ps(rsq00,rinv00);
 381             r00              = _mm_andnot_ps(dummy_mask,r00);
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq00             = _mm_mul_ps(iq0,jq0);
 385             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 386                                          vdwparam+vdwioffset0+vdwjidx0B,
 387                                          vdwparam+vdwioffset0+vdwjidx0C,
 388                                          vdwparam+vdwioffset0+vdwjidx0D,
 389                                          &c6_00,&c12_00);
 390
 391             /* Calculate table index by multiplying r with table scale and truncate to integer */
 392             rt               = _mm_mul_ps(r00,vftabscale);
 393             vfitab           = _mm_cvttps_epi32(rt);
 394             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 395             vfitab           = _mm_slli_epi32(vfitab,3);
 396
 397             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 398             isaprod          = _mm_mul_ps(isai0,isaj0);
 399             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 400             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 401
 402             /* Calculate generalized born table index - this is a separate table from the normal one,
 403              * but we use the same procedure by multiplying r with scale and truncating to integer.
 404              */
 405             rt               = _mm_mul_ps(r00,gbscale);
 406             gbitab           = _mm_cvttps_epi32(rt);
 407             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 408             gbitab           = _mm_slli_epi32(gbitab,2);
 409
 410             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 411             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 412             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 413             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 414             _MM_TRANSPOSE4_PS(Y,F,G,H);
 415             Heps             = _mm_mul_ps(gbeps,H);
 416             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 417             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 418             vgb              = _mm_mul_ps(gbqqfactor,VV);
 419
 420             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 421             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 422             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 423             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 424             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 425             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 426             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 427             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 428             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 429             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 430             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 431             velec            = _mm_mul_ps(qq00,rinv00);
 432             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 433
 434             /* CUBIC SPLINE TABLE DISPERSION */
 435             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 436             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 437             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 438             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 439             _MM_TRANSPOSE4_PS(Y,F,G,H);
 440             Heps             = _mm_mul_ps(vfeps,H);
 441             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 442             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 443             vvdw6            = _mm_mul_ps(c6_00,VV);
 444             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 445             fvdw6            = _mm_mul_ps(c6_00,FF);
 446
 447             /* CUBIC SPLINE TABLE REPULSION */
 448             vfitab           = _mm_add_epi32(vfitab,ifour);
 449             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 450             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 451             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 452             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 453             _MM_TRANSPOSE4_PS(Y,F,G,H);
 454             Heps             = _mm_mul_ps(vfeps,H);
 455             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 456             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 457             vvdw12           = _mm_mul_ps(c12_00,VV);
 458             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 459             fvdw12           = _mm_mul_ps(c12_00,FF);
 460             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 461             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 462
 463             /* Update potential sum for this i atom from the interaction with this j atom. */
 464             velec            = _mm_andnot_ps(dummy_mask,velec);
 465             velecsum         = _mm_add_ps(velecsum,velec);
 466             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 467             vgbsum           = _mm_add_ps(vgbsum,vgb);
 468             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 469             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 470
 471             fscal            = _mm_add_ps(felec,fvdw);
 472
 473             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 474
 475             /* Calculate temporary vectorial force */
 476             tx               = _mm_mul_ps(fscal,dx00);
 477             ty               = _mm_mul_ps(fscal,dy00);
 478             tz               = _mm_mul_ps(fscal,dz00);
 479
 480             /* Update vectorial force */
 481             fix0             = _mm_add_ps(fix0,tx);
 482             fiy0             = _mm_add_ps(fiy0,ty);
 483             fiz0             = _mm_add_ps(fiz0,tz);
 484
 485             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 486             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 487             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 488             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 489             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 490
 491             /* Inner loop uses 93 flops */
 492         }
 493
 494         /* End of innermost loop */
 495
 496         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 497                                               f+i_coord_offset,fshift+i_shift_offset);
 498
 499         ggid                        = gid[iidx];
 500         /* Update potential energies */
 501         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 502         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 503         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 504         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 505         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 506
 507         /* Increment number of inner iterations */
 508         inneriter                  += j_index_end - j_index_start;
 509
 510         /* Outer loop uses 10 flops */
 511     }
 512
 513     /* Increment number of outer iterations */
 514     outeriter        += nri;
 515
 516     /* Update outer/inner flops */
 517
 518     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*93);
 519 }
 520 /*
 521  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_single
 522  * Electrostatics interaction: GeneralizedBorn
 523  * VdW interaction:            CubicSplineTable
 524  * Geometry:                   Particle-Particle
 525  * Calculate force/pot:        Force
 526  */
 527 void
 528 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_single
 529                     (t_nblist                    * gmx_restrict       nlist,
 530                      rvec                        * gmx_restrict          xx,
 531                      rvec                        * gmx_restrict          ff,
 532                      struct t_forcerec           * gmx_restrict          fr,
 533                      t_mdatoms                   * gmx_restrict     mdatoms,
 534                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 535                      t_nrnb                      * gmx_restrict        nrnb)
 536 {
 537     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 538      * just 0 for non-waters.
 539      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 540      * jnr indices corresponding to data put in the four positions in the SIMD register.
 541      */
 542     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 543     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 544     int              jnrA,jnrB,jnrC,jnrD;
 545     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 546     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 547     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 548     real             rcutoff_scalar;
 549     real             *shiftvec,*fshift,*x,*f;
 550     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 551     real             scratch[4*DIM];
 552     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 553     int              vdwioffset0;
 554     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 555     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 556     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 557     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 558     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 559     real             *charge;
 560     __m128i          gbitab;
 561     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 562     __m128           minushalf = _mm_set1_ps(-0.5);
 563     real             *invsqrta,*dvda,*gbtab;
 564     int              nvdwtype;
 565     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 566     int              *vdwtype;
 567     real             *vdwparam;
 568     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 569     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 570     __m128i          vfitab;
 571     __m128i          ifour       = _mm_set1_epi32(4);
 572     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 573     real             *vftab;
 574     __m128           dummy_mask,cutoff_mask;
 575     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 576     __m128           one     = _mm_set1_ps(1.0);
 577     __m128           two     = _mm_set1_ps(2.0);
 578     x                = xx[0];
 579     f                = ff[0];
 580
 581     nri              = nlist->nri;
 582     iinr             = nlist->iinr;
 583     jindex           = nlist->jindex;
 584     jjnr             = nlist->jjnr;
 585     shiftidx         = nlist->shift;
 586     gid              = nlist->gid;
 587     shiftvec         = fr->shift_vec[0];
 588     fshift           = fr->fshift[0];
 589     facel            = _mm_set1_ps(fr->ic->epsfac);
 590     charge           = mdatoms->chargeA;
 591     nvdwtype         = fr->ntype;
 592     vdwparam         = fr->nbfp;
 593     vdwtype          = mdatoms->typeA;
 594
 595     vftab            = kernel_data->table_vdw->data;
 596     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 597
 598     invsqrta         = fr->invsqrta;
 599     dvda             = fr->dvda;
 600     gbtabscale       = _mm_set1_ps(fr->gbtab->scale);
 601     gbtab            = fr->gbtab->data;
 602     gbinvepsdiff     = _mm_set1_ps((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 603
 604     /* Avoid stupid compiler warnings */
 605     jnrA = jnrB = jnrC = jnrD = 0;
 606     j_coord_offsetA = 0;
 607     j_coord_offsetB = 0;
 608     j_coord_offsetC = 0;
 609     j_coord_offsetD = 0;
 610
 611     outeriter        = 0;
 612     inneriter        = 0;
 613
 614     for(iidx=0;iidx<4*DIM;iidx++)
 615     {
 616         scratch[iidx] = 0.0;
 617     }
 618
 619     /* Start outer loop over neighborlists */
 620     for(iidx=0; iidx<nri; iidx++)
 621     {
 622         /* Load shift vector for this list */
 623         i_shift_offset   = DIM*shiftidx[iidx];
 624
 625         /* Load limits for loop over neighbors */
 626         j_index_start    = jindex[iidx];
 627         j_index_end      = jindex[iidx+1];
 628
 629         /* Get outer coordinate index */
 630         inr              = iinr[iidx];
 631         i_coord_offset   = DIM*inr;
 632
 633         /* Load i particle coords and add shift vector */
 634         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 635
 636         fix0             = _mm_setzero_ps();
 637         fiy0             = _mm_setzero_ps();
 638         fiz0             = _mm_setzero_ps();
 639
 640         /* Load parameters for i particles */
 641         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 642         isai0            = _mm_load1_ps(invsqrta+inr+0);
 643         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 644
 645         dvdasum          = _mm_setzero_ps();
 646
 647         /* Start inner kernel loop */
 648         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 649         {
 650
 651             /* Get j neighbor index, and coordinate index */
 652             jnrA             = jjnr[jidx];
 653             jnrB             = jjnr[jidx+1];
 654             jnrC             = jjnr[jidx+2];
 655             jnrD             = jjnr[jidx+3];
 656             j_coord_offsetA  = DIM*jnrA;
 657             j_coord_offsetB  = DIM*jnrB;
 658             j_coord_offsetC  = DIM*jnrC;
 659             j_coord_offsetD  = DIM*jnrD;
 660
 661             /* load j atom coordinates */
 662             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 663                                               x+j_coord_offsetC,x+j_coord_offsetD,
 664                                               &jx0,&jy0,&jz0);
 665
 666             /* Calculate displacement vector */
 667             dx00             = _mm_sub_ps(ix0,jx0);
 668             dy00             = _mm_sub_ps(iy0,jy0);
 669             dz00             = _mm_sub_ps(iz0,jz0);
 670
 671             /* Calculate squared distance and things based on it */
 672             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 673
 674             rinv00           = sse2_invsqrt_f(rsq00);
 675
 676             /* Load parameters for j particles */
 677             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 678                                                               charge+jnrC+0,charge+jnrD+0);
 679             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 680                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 681             vdwjidx0A        = 2*vdwtype[jnrA+0];
 682             vdwjidx0B        = 2*vdwtype[jnrB+0];
 683             vdwjidx0C        = 2*vdwtype[jnrC+0];
 684             vdwjidx0D        = 2*vdwtype[jnrD+0];
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             r00              = _mm_mul_ps(rsq00,rinv00);
 691
 692             /* Compute parameters for interactions between i and j atoms */
 693             qq00             = _mm_mul_ps(iq0,jq0);
 694             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 695                                          vdwparam+vdwioffset0+vdwjidx0B,
 696                                          vdwparam+vdwioffset0+vdwjidx0C,
 697                                          vdwparam+vdwioffset0+vdwjidx0D,
 698                                          &c6_00,&c12_00);
 699
 700             /* Calculate table index by multiplying r with table scale and truncate to integer */
 701             rt               = _mm_mul_ps(r00,vftabscale);
 702             vfitab           = _mm_cvttps_epi32(rt);
 703             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 704             vfitab           = _mm_slli_epi32(vfitab,3);
 705
 706             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 707             isaprod          = _mm_mul_ps(isai0,isaj0);
 708             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 709             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 710
 711             /* Calculate generalized born table index - this is a separate table from the normal one,
 712              * but we use the same procedure by multiplying r with scale and truncating to integer.
 713              */
 714             rt               = _mm_mul_ps(r00,gbscale);
 715             gbitab           = _mm_cvttps_epi32(rt);
 716             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 717             gbitab           = _mm_slli_epi32(gbitab,2);
 718
 719             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 720             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 721             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 722             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 723             _MM_TRANSPOSE4_PS(Y,F,G,H);
 724             Heps             = _mm_mul_ps(gbeps,H);
 725             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 726             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 727             vgb              = _mm_mul_ps(gbqqfactor,VV);
 728
 729             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 730             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 731             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 732             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 733             fjptrA           = dvda+jnrA;
 734             fjptrB           = dvda+jnrB;
 735             fjptrC           = dvda+jnrC;
 736             fjptrD           = dvda+jnrD;
 737             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 738             velec            = _mm_mul_ps(qq00,rinv00);
 739             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 740
 741             /* CUBIC SPLINE TABLE DISPERSION */
 742             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 743             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 744             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 745             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 746             _MM_TRANSPOSE4_PS(Y,F,G,H);
 747             Heps             = _mm_mul_ps(vfeps,H);
 748             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 749             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 750             fvdw6            = _mm_mul_ps(c6_00,FF);
 751
 752             /* CUBIC SPLINE TABLE REPULSION */
 753             vfitab           = _mm_add_epi32(vfitab,ifour);
 754             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 755             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 756             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 757             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 758             _MM_TRANSPOSE4_PS(Y,F,G,H);
 759             Heps             = _mm_mul_ps(vfeps,H);
 760             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 761             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 762             fvdw12           = _mm_mul_ps(c12_00,FF);
 763             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 764
 765             fscal            = _mm_add_ps(felec,fvdw);
 766
 767             /* Calculate temporary vectorial force */
 768             tx               = _mm_mul_ps(fscal,dx00);
 769             ty               = _mm_mul_ps(fscal,dy00);
 770             tz               = _mm_mul_ps(fscal,dz00);
 771
 772             /* Update vectorial force */
 773             fix0             = _mm_add_ps(fix0,tx);
 774             fiy0             = _mm_add_ps(fiy0,ty);
 775             fiz0             = _mm_add_ps(fiz0,tz);
 776
 777             fjptrA             = f+j_coord_offsetA;
 778             fjptrB             = f+j_coord_offsetB;
 779             fjptrC             = f+j_coord_offsetC;
 780             fjptrD             = f+j_coord_offsetD;
 781             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 782
 783             /* Inner loop uses 82 flops */
 784         }
 785
 786         if(jidx<j_index_end)
 787         {
 788
 789             /* Get j neighbor index, and coordinate index */
 790             jnrlistA         = jjnr[jidx];
 791             jnrlistB         = jjnr[jidx+1];
 792             jnrlistC         = jjnr[jidx+2];
 793             jnrlistD         = jjnr[jidx+3];
 794             /* Sign of each element will be negative for non-real atoms.
 795              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 796              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 797              */
 798             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 799             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 800             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 801             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 802             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 803             j_coord_offsetA  = DIM*jnrA;
 804             j_coord_offsetB  = DIM*jnrB;
 805             j_coord_offsetC  = DIM*jnrC;
 806             j_coord_offsetD  = DIM*jnrD;
 807
 808             /* load j atom coordinates */
 809             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 810                                               x+j_coord_offsetC,x+j_coord_offsetD,
 811                                               &jx0,&jy0,&jz0);
 812
 813             /* Calculate displacement vector */
 814             dx00             = _mm_sub_ps(ix0,jx0);
 815             dy00             = _mm_sub_ps(iy0,jy0);
 816             dz00             = _mm_sub_ps(iz0,jz0);
 817
 818             /* Calculate squared distance and things based on it */
 819             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 820
 821             rinv00           = sse2_invsqrt_f(rsq00);
 822
 823             /* Load parameters for j particles */
 824             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 825                                                               charge+jnrC+0,charge+jnrD+0);
 826             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 827                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 828             vdwjidx0A        = 2*vdwtype[jnrA+0];
 829             vdwjidx0B        = 2*vdwtype[jnrB+0];
 830             vdwjidx0C        = 2*vdwtype[jnrC+0];
 831             vdwjidx0D        = 2*vdwtype[jnrD+0];
 832
 833             /**************************
 834              * CALCULATE INTERACTIONS *
 835              **************************/
 836
 837             r00              = _mm_mul_ps(rsq00,rinv00);
 838             r00              = _mm_andnot_ps(dummy_mask,r00);
 839
 840             /* Compute parameters for interactions between i and j atoms */
 841             qq00             = _mm_mul_ps(iq0,jq0);
 842             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 843                                          vdwparam+vdwioffset0+vdwjidx0B,
 844                                          vdwparam+vdwioffset0+vdwjidx0C,
 845                                          vdwparam+vdwioffset0+vdwjidx0D,
 846                                          &c6_00,&c12_00);
 847
 848             /* Calculate table index by multiplying r with table scale and truncate to integer */
 849             rt               = _mm_mul_ps(r00,vftabscale);
 850             vfitab           = _mm_cvttps_epi32(rt);
 851             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 852             vfitab           = _mm_slli_epi32(vfitab,3);
 853
 854             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 855             isaprod          = _mm_mul_ps(isai0,isaj0);
 856             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 857             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 858
 859             /* Calculate generalized born table index - this is a separate table from the normal one,
 860              * but we use the same procedure by multiplying r with scale and truncating to integer.
 861              */
 862             rt               = _mm_mul_ps(r00,gbscale);
 863             gbitab           = _mm_cvttps_epi32(rt);
 864             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 865             gbitab           = _mm_slli_epi32(gbitab,2);
 866
 867             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 868             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 869             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 870             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 871             _MM_TRANSPOSE4_PS(Y,F,G,H);
 872             Heps             = _mm_mul_ps(gbeps,H);
 873             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 874             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 875             vgb              = _mm_mul_ps(gbqqfactor,VV);
 876
 877             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 878             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 879             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 880             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 881             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 882             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 883             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 884             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 885             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 886             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 887             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 888             velec            = _mm_mul_ps(qq00,rinv00);
 889             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 890
 891             /* CUBIC SPLINE TABLE DISPERSION */
 892             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 893             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 894             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 895             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 896             _MM_TRANSPOSE4_PS(Y,F,G,H);
 897             Heps             = _mm_mul_ps(vfeps,H);
 898             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 899             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 900             fvdw6            = _mm_mul_ps(c6_00,FF);
 901
 902             /* CUBIC SPLINE TABLE REPULSION */
 903             vfitab           = _mm_add_epi32(vfitab,ifour);
 904             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 905             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 906             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 907             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 908             _MM_TRANSPOSE4_PS(Y,F,G,H);
 909             Heps             = _mm_mul_ps(vfeps,H);
 910             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 911             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 912             fvdw12           = _mm_mul_ps(c12_00,FF);
 913             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 914
 915             fscal            = _mm_add_ps(felec,fvdw);
 916
 917             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 918
 919             /* Calculate temporary vectorial force */
 920             tx               = _mm_mul_ps(fscal,dx00);
 921             ty               = _mm_mul_ps(fscal,dy00);
 922             tz               = _mm_mul_ps(fscal,dz00);
 923
 924             /* Update vectorial force */
 925             fix0             = _mm_add_ps(fix0,tx);
 926             fiy0             = _mm_add_ps(fiy0,ty);
 927             fiz0             = _mm_add_ps(fiz0,tz);
 928
 929             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 930             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 931             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 932             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 933             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 934
 935             /* Inner loop uses 83 flops */
 936         }
 937
 938         /* End of innermost loop */
 939
 940         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 941                                               f+i_coord_offset,fshift+i_shift_offset);
 942
 943         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 944         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 945
 946         /* Increment number of inner iterations */
 947         inneriter                  += j_index_end - j_index_start;
 948
 949         /* Outer loop uses 7 flops */
 950     }
 951
 952     /* Increment number of outer iterations */
 953     outeriter        += nri;
 954
 955     /* Update outer/inner flops */
 956
 957     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);
 958 }