src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecGB_VdwCSTab_GeomP1P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012, by the GROMACS development team, led by
   5  * David van der Spoel, Berk Hess, Erik Lindahl, and including many
   6  * others, as listed in the AUTHORS file in the top-level source
   7  * directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: GeneralizedBorn
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse4_1_single
  61                     (t_nblist * gmx_restrict                nlist,
  62                      rvec * gmx_restrict                    xx,
  63                      rvec * gmx_restrict                    ff,
  64                      t_forcerec * gmx_restrict              fr,
  65                      t_mdatoms * gmx_restrict               mdatoms,
  66                      nb_kernel_data_t * gmx_restrict        kernel_data,
  67                      t_nrnb * gmx_restrict                  nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128i          gbitab;
  93     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  94     __m128           minushalf = _mm_set1_ps(-0.5);
  95     real             *invsqrta,*dvda,*gbtab;
  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          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_vdw->data;
 128     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 129
 130     invsqrta         = fr->invsqrta;
 131     dvda             = fr->dvda;
 132     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 133     gbtab            = fr->gbtab.data;
 134     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     for(iidx=0;iidx<4*DIM;iidx++)
 147     {
 148         scratch[iidx] = 0.0;
 149     }
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 167
 168         fix0             = _mm_setzero_ps();
 169         fiy0             = _mm_setzero_ps();
 170         fiz0             = _mm_setzero_ps();
 171
 172         /* Load parameters for i particles */
 173         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 174         isai0            = _mm_load1_ps(invsqrta+inr+0);
 175         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm_setzero_ps();
 179         vgbsum           = _mm_setzero_ps();
 180         vvdwsum          = _mm_setzero_ps();
 181         dvdasum          = _mm_setzero_ps();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               x+j_coord_offsetC,x+j_coord_offsetD,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_ps(ix0,jx0);
 204             dy00             = _mm_sub_ps(iy0,jy0);
 205             dz00             = _mm_sub_ps(iz0,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 209
 210             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 214                                                               charge+jnrC+0,charge+jnrD+0);
 215             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 216                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 217             vdwjidx0A        = 2*vdwtype[jnrA+0];
 218             vdwjidx0B        = 2*vdwtype[jnrB+0];
 219             vdwjidx0C        = 2*vdwtype[jnrC+0];
 220             vdwjidx0D        = 2*vdwtype[jnrD+0];
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             r00              = _mm_mul_ps(rsq00,rinv00);
 227
 228             /* Compute parameters for interactions between i and j atoms */
 229             qq00             = _mm_mul_ps(iq0,jq0);
 230             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 231                                          vdwparam+vdwioffset0+vdwjidx0B,
 232                                          vdwparam+vdwioffset0+vdwjidx0C,
 233                                          vdwparam+vdwioffset0+vdwjidx0D,
 234                                          &c6_00,&c12_00);
 235
 236             /* Calculate table index by multiplying r with table scale and truncate to integer */
 237             rt               = _mm_mul_ps(r00,vftabscale);
 238             vfitab           = _mm_cvttps_epi32(rt);
 239             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 240             vfitab           = _mm_slli_epi32(vfitab,3);
 241
 242             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 243             isaprod          = _mm_mul_ps(isai0,isaj0);
 244             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 245             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 246
 247             /* Calculate generalized born table index - this is a separate table from the normal one,
 248              * but we use the same procedure by multiplying r with scale and truncating to integer.
 249              */
 250             rt               = _mm_mul_ps(r00,gbscale);
 251             gbitab           = _mm_cvttps_epi32(rt);
 252             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 253             gbitab           = _mm_slli_epi32(gbitab,2);
 254             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 255             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 256             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 257             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 258             _MM_TRANSPOSE4_PS(Y,F,G,H);
 259             Heps             = _mm_mul_ps(gbeps,H);
 260             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 261             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 262             vgb              = _mm_mul_ps(gbqqfactor,VV);
 263
 264             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 265             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 266             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 267             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 268             fjptrA           = dvda+jnrA;
 269             fjptrB           = dvda+jnrB;
 270             fjptrC           = dvda+jnrC;
 271             fjptrD           = dvda+jnrD;
 272             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 273             velec            = _mm_mul_ps(qq00,rinv00);
 274             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 275
 276             /* CUBIC SPLINE TABLE DISPERSION */
 277             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 278             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 279             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 280             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 281             _MM_TRANSPOSE4_PS(Y,F,G,H);
 282             Heps             = _mm_mul_ps(vfeps,H);
 283             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 284             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 285             vvdw6            = _mm_mul_ps(c6_00,VV);
 286             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 287             fvdw6            = _mm_mul_ps(c6_00,FF);
 288
 289             /* CUBIC SPLINE TABLE REPULSION */
 290             vfitab           = _mm_add_epi32(vfitab,ifour);
 291             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 292             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 293             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 294             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 295             _MM_TRANSPOSE4_PS(Y,F,G,H);
 296             Heps             = _mm_mul_ps(vfeps,H);
 297             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 298             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 299             vvdw12           = _mm_mul_ps(c12_00,VV);
 300             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 301             fvdw12           = _mm_mul_ps(c12_00,FF);
 302             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 303             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             velecsum         = _mm_add_ps(velecsum,velec);
 307             vgbsum           = _mm_add_ps(vgbsum,vgb);
 308             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 309
 310             fscal            = _mm_add_ps(felec,fvdw);
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm_mul_ps(fscal,dx00);
 314             ty               = _mm_mul_ps(fscal,dy00);
 315             tz               = _mm_mul_ps(fscal,dz00);
 316
 317             /* Update vectorial force */
 318             fix0             = _mm_add_ps(fix0,tx);
 319             fiy0             = _mm_add_ps(fiy0,ty);
 320             fiz0             = _mm_add_ps(fiz0,tz);
 321
 322             fjptrA             = f+j_coord_offsetA;
 323             fjptrB             = f+j_coord_offsetB;
 324             fjptrC             = f+j_coord_offsetC;
 325             fjptrD             = f+j_coord_offsetD;
 326             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 327
 328             /* Inner loop uses 92 flops */
 329         }
 330
 331         if(jidx<j_index_end)
 332         {
 333
 334             /* Get j neighbor index, and coordinate index */
 335             jnrlistA         = jjnr[jidx];
 336             jnrlistB         = jjnr[jidx+1];
 337             jnrlistC         = jjnr[jidx+2];
 338             jnrlistD         = jjnr[jidx+3];
 339             /* Sign of each element will be negative for non-real atoms.
 340              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 341              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 342              */
 343             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 344             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 345             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 346             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 347             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 348             j_coord_offsetA  = DIM*jnrA;
 349             j_coord_offsetB  = DIM*jnrB;
 350             j_coord_offsetC  = DIM*jnrC;
 351             j_coord_offsetD  = DIM*jnrD;
 352
 353             /* load j atom coordinates */
 354             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 355                                               x+j_coord_offsetC,x+j_coord_offsetD,
 356                                               &jx0,&jy0,&jz0);
 357
 358             /* Calculate displacement vector */
 359             dx00             = _mm_sub_ps(ix0,jx0);
 360             dy00             = _mm_sub_ps(iy0,jy0);
 361             dz00             = _mm_sub_ps(iz0,jz0);
 362
 363             /* Calculate squared distance and things based on it */
 364             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 365
 366             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 367
 368             /* Load parameters for j particles */
 369             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 370                                                               charge+jnrC+0,charge+jnrD+0);
 371             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 372                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 373             vdwjidx0A        = 2*vdwtype[jnrA+0];
 374             vdwjidx0B        = 2*vdwtype[jnrB+0];
 375             vdwjidx0C        = 2*vdwtype[jnrC+0];
 376             vdwjidx0D        = 2*vdwtype[jnrD+0];
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             r00              = _mm_mul_ps(rsq00,rinv00);
 383             r00              = _mm_andnot_ps(dummy_mask,r00);
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq00             = _mm_mul_ps(iq0,jq0);
 387             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 388                                          vdwparam+vdwioffset0+vdwjidx0B,
 389                                          vdwparam+vdwioffset0+vdwjidx0C,
 390                                          vdwparam+vdwioffset0+vdwjidx0D,
 391                                          &c6_00,&c12_00);
 392
 393             /* Calculate table index by multiplying r with table scale and truncate to integer */
 394             rt               = _mm_mul_ps(r00,vftabscale);
 395             vfitab           = _mm_cvttps_epi32(rt);
 396             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 397             vfitab           = _mm_slli_epi32(vfitab,3);
 398
 399             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 400             isaprod          = _mm_mul_ps(isai0,isaj0);
 401             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 402             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 403
 404             /* Calculate generalized born table index - this is a separate table from the normal one,
 405              * but we use the same procedure by multiplying r with scale and truncating to integer.
 406              */
 407             rt               = _mm_mul_ps(r00,gbscale);
 408             gbitab           = _mm_cvttps_epi32(rt);
 409             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 410             gbitab           = _mm_slli_epi32(gbitab,2);
 411             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 412             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 413             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 414             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 415             _MM_TRANSPOSE4_PS(Y,F,G,H);
 416             Heps             = _mm_mul_ps(gbeps,H);
 417             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 418             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 419             vgb              = _mm_mul_ps(gbqqfactor,VV);
 420
 421             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 422             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 423             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 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_sse4_1_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_sse4_1_single
 529                     (t_nblist * gmx_restrict                nlist,
 530                      rvec * gmx_restrict                    xx,
 531                      rvec * gmx_restrict                    ff,
 532                      t_forcerec * gmx_restrict              fr,
 533                      t_mdatoms * gmx_restrict               mdatoms,
 534                      nb_kernel_data_t * 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->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->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           = gmx_mm_invsqrt_ps(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_round_ps(rt, _MM_FROUND_FLOOR));
 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_round_ps(rt, _MM_FROUND_FLOOR));
 717             gbitab           = _mm_slli_epi32(gbitab,2);
 718             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 719             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 720             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 721             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 722             _MM_TRANSPOSE4_PS(Y,F,G,H);
 723             Heps             = _mm_mul_ps(gbeps,H);
 724             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 725             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 726             vgb              = _mm_mul_ps(gbqqfactor,VV);
 727
 728             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 729             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 730             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 731             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 732             fjptrA           = dvda+jnrA;
 733             fjptrB           = dvda+jnrB;
 734             fjptrC           = dvda+jnrC;
 735             fjptrD           = dvda+jnrD;
 736             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 737             velec            = _mm_mul_ps(qq00,rinv00);
 738             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 739
 740             /* CUBIC SPLINE TABLE DISPERSION */
 741             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 742             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 743             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 744             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 745             _MM_TRANSPOSE4_PS(Y,F,G,H);
 746             Heps             = _mm_mul_ps(vfeps,H);
 747             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 748             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 749             fvdw6            = _mm_mul_ps(c6_00,FF);
 750
 751             /* CUBIC SPLINE TABLE REPULSION */
 752             vfitab           = _mm_add_epi32(vfitab,ifour);
 753             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 754             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 755             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 756             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 757             _MM_TRANSPOSE4_PS(Y,F,G,H);
 758             Heps             = _mm_mul_ps(vfeps,H);
 759             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 760             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 761             fvdw12           = _mm_mul_ps(c12_00,FF);
 762             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 763
 764             fscal            = _mm_add_ps(felec,fvdw);
 765
 766             /* Calculate temporary vectorial force */
 767             tx               = _mm_mul_ps(fscal,dx00);
 768             ty               = _mm_mul_ps(fscal,dy00);
 769             tz               = _mm_mul_ps(fscal,dz00);
 770
 771             /* Update vectorial force */
 772             fix0             = _mm_add_ps(fix0,tx);
 773             fiy0             = _mm_add_ps(fiy0,ty);
 774             fiz0             = _mm_add_ps(fiz0,tz);
 775
 776             fjptrA             = f+j_coord_offsetA;
 777             fjptrB             = f+j_coord_offsetB;
 778             fjptrC             = f+j_coord_offsetC;
 779             fjptrD             = f+j_coord_offsetD;
 780             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 781
 782             /* Inner loop uses 82 flops */
 783         }
 784
 785         if(jidx<j_index_end)
 786         {
 787
 788             /* Get j neighbor index, and coordinate index */
 789             jnrlistA         = jjnr[jidx];
 790             jnrlistB         = jjnr[jidx+1];
 791             jnrlistC         = jjnr[jidx+2];
 792             jnrlistD         = jjnr[jidx+3];
 793             /* Sign of each element will be negative for non-real atoms.
 794              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 795              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 796              */
 797             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 798             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 799             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 800             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 801             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 802             j_coord_offsetA  = DIM*jnrA;
 803             j_coord_offsetB  = DIM*jnrB;
 804             j_coord_offsetC  = DIM*jnrC;
 805             j_coord_offsetD  = DIM*jnrD;
 806
 807             /* load j atom coordinates */
 808             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 809                                               x+j_coord_offsetC,x+j_coord_offsetD,
 810                                               &jx0,&jy0,&jz0);
 811
 812             /* Calculate displacement vector */
 813             dx00             = _mm_sub_ps(ix0,jx0);
 814             dy00             = _mm_sub_ps(iy0,jy0);
 815             dz00             = _mm_sub_ps(iz0,jz0);
 816
 817             /* Calculate squared distance and things based on it */
 818             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 819
 820             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 821
 822             /* Load parameters for j particles */
 823             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 824                                                               charge+jnrC+0,charge+jnrD+0);
 825             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 826                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 827             vdwjidx0A        = 2*vdwtype[jnrA+0];
 828             vdwjidx0B        = 2*vdwtype[jnrB+0];
 829             vdwjidx0C        = 2*vdwtype[jnrC+0];
 830             vdwjidx0D        = 2*vdwtype[jnrD+0];
 831
 832             /**************************
 833              * CALCULATE INTERACTIONS *
 834              **************************/
 835
 836             r00              = _mm_mul_ps(rsq00,rinv00);
 837             r00              = _mm_andnot_ps(dummy_mask,r00);
 838
 839             /* Compute parameters for interactions between i and j atoms */
 840             qq00             = _mm_mul_ps(iq0,jq0);
 841             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 842                                          vdwparam+vdwioffset0+vdwjidx0B,
 843                                          vdwparam+vdwioffset0+vdwjidx0C,
 844                                          vdwparam+vdwioffset0+vdwjidx0D,
 845                                          &c6_00,&c12_00);
 846
 847             /* Calculate table index by multiplying r with table scale and truncate to integer */
 848             rt               = _mm_mul_ps(r00,vftabscale);
 849             vfitab           = _mm_cvttps_epi32(rt);
 850             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 851             vfitab           = _mm_slli_epi32(vfitab,3);
 852
 853             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 854             isaprod          = _mm_mul_ps(isai0,isaj0);
 855             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 856             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 857
 858             /* Calculate generalized born table index - this is a separate table from the normal one,
 859              * but we use the same procedure by multiplying r with scale and truncating to integer.
 860              */
 861             rt               = _mm_mul_ps(r00,gbscale);
 862             gbitab           = _mm_cvttps_epi32(rt);
 863             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 864             gbitab           = _mm_slli_epi32(gbitab,2);
 865             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 866             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 867             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 868             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 869             _MM_TRANSPOSE4_PS(Y,F,G,H);
 870             Heps             = _mm_mul_ps(gbeps,H);
 871             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 872             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 873             vgb              = _mm_mul_ps(gbqqfactor,VV);
 874
 875             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 876             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 877             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 878             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 879             /* 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. */
 880             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 881             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 882             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 883             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 884             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 885             velec            = _mm_mul_ps(qq00,rinv00);
 886             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 887
 888             /* CUBIC SPLINE TABLE DISPERSION */
 889             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 890             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 891             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 892             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 893             _MM_TRANSPOSE4_PS(Y,F,G,H);
 894             Heps             = _mm_mul_ps(vfeps,H);
 895             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 896             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 897             fvdw6            = _mm_mul_ps(c6_00,FF);
 898
 899             /* CUBIC SPLINE TABLE REPULSION */
 900             vfitab           = _mm_add_epi32(vfitab,ifour);
 901             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 902             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 903             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 904             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 905             _MM_TRANSPOSE4_PS(Y,F,G,H);
 906             Heps             = _mm_mul_ps(vfeps,H);
 907             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 908             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 909             fvdw12           = _mm_mul_ps(c12_00,FF);
 910             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 911
 912             fscal            = _mm_add_ps(felec,fvdw);
 913
 914             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 915
 916             /* Calculate temporary vectorial force */
 917             tx               = _mm_mul_ps(fscal,dx00);
 918             ty               = _mm_mul_ps(fscal,dy00);
 919             tz               = _mm_mul_ps(fscal,dz00);
 920
 921             /* Update vectorial force */
 922             fix0             = _mm_add_ps(fix0,tx);
 923             fiy0             = _mm_add_ps(fiy0,ty);
 924             fiz0             = _mm_add_ps(fiz0,tz);
 925
 926             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 927             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 928             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 929             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 930             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 931
 932             /* Inner loop uses 83 flops */
 933         }
 934
 935         /* End of innermost loop */
 936
 937         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 938                                               f+i_coord_offset,fshift+i_shift_offset);
 939
 940         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 941         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 942
 943         /* Increment number of inner iterations */
 944         inneriter                  += j_index_end - j_index_start;
 945
 946         /* Outer loop uses 7 flops */
 947     }
 948
 949     /* Increment number of outer iterations */
 950     outeriter        += nri;
 951
 952     /* Update outer/inner flops */
 953
 954     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);
 955 }