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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_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_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              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_cvtepi32_ps(vfitab));
 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_cvtepi32_ps(gbitab));
 253             gbitab           = _mm_slli_epi32(gbitab,2);
 254
 255             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 256             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 257             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 258             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 259             _MM_TRANSPOSE4_PS(Y,F,G,H);
 260             Heps             = _mm_mul_ps(gbeps,H);
 261             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 262             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 263             vgb              = _mm_mul_ps(gbqqfactor,VV);
 264
 265             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 266             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 267             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 268             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 269             fjptrA           = dvda+jnrA;
 270             fjptrB           = dvda+jnrB;
 271             fjptrC           = dvda+jnrC;
 272             fjptrD           = dvda+jnrD;
 273             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 274             velec            = _mm_mul_ps(qq00,rinv00);
 275             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 276
 277             /* CUBIC SPLINE TABLE DISPERSION */
 278             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 279             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 280             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 281             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 282             _MM_TRANSPOSE4_PS(Y,F,G,H);
 283             Heps             = _mm_mul_ps(vfeps,H);
 284             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 285             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 286             vvdw6            = _mm_mul_ps(c6_00,VV);
 287             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 288             fvdw6            = _mm_mul_ps(c6_00,FF);
 289
 290             /* CUBIC SPLINE TABLE REPULSION */
 291             vfitab           = _mm_add_epi32(vfitab,ifour);
 292             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 293             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 294             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 295             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 296             _MM_TRANSPOSE4_PS(Y,F,G,H);
 297             Heps             = _mm_mul_ps(vfeps,H);
 298             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 299             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 300             vvdw12           = _mm_mul_ps(c12_00,VV);
 301             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 302             fvdw12           = _mm_mul_ps(c12_00,FF);
 303             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 304             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velecsum         = _mm_add_ps(velecsum,velec);
 308             vgbsum           = _mm_add_ps(vgbsum,vgb);
 309             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 310
 311             fscal            = _mm_add_ps(felec,fvdw);
 312
 313             /* Calculate temporary vectorial force */
 314             tx               = _mm_mul_ps(fscal,dx00);
 315             ty               = _mm_mul_ps(fscal,dy00);
 316             tz               = _mm_mul_ps(fscal,dz00);
 317
 318             /* Update vectorial force */
 319             fix0             = _mm_add_ps(fix0,tx);
 320             fiy0             = _mm_add_ps(fiy0,ty);
 321             fiz0             = _mm_add_ps(fiz0,tz);
 322
 323             fjptrA             = f+j_coord_offsetA;
 324             fjptrB             = f+j_coord_offsetB;
 325             fjptrC             = f+j_coord_offsetC;
 326             fjptrD             = f+j_coord_offsetD;
 327             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 328
 329             /* Inner loop uses 92 flops */
 330         }
 331
 332         if(jidx<j_index_end)
 333         {
 334
 335             /* Get j neighbor index, and coordinate index */
 336             jnrlistA         = jjnr[jidx];
 337             jnrlistB         = jjnr[jidx+1];
 338             jnrlistC         = jjnr[jidx+2];
 339             jnrlistD         = jjnr[jidx+3];
 340             /* Sign of each element will be negative for non-real atoms.
 341              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 342              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 343              */
 344             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 345             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 346             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 347             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 348             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 349             j_coord_offsetA  = DIM*jnrA;
 350             j_coord_offsetB  = DIM*jnrB;
 351             j_coord_offsetC  = DIM*jnrC;
 352             j_coord_offsetD  = DIM*jnrD;
 353
 354             /* load j atom coordinates */
 355             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 356                                               x+j_coord_offsetC,x+j_coord_offsetD,
 357                                               &jx0,&jy0,&jz0);
 358
 359             /* Calculate displacement vector */
 360             dx00             = _mm_sub_ps(ix0,jx0);
 361             dy00             = _mm_sub_ps(iy0,jy0);
 362             dz00             = _mm_sub_ps(iz0,jz0);
 363
 364             /* Calculate squared distance and things based on it */
 365             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 366
 367             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 368
 369             /* Load parameters for j particles */
 370             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 371                                                               charge+jnrC+0,charge+jnrD+0);
 372             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 373                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 374             vdwjidx0A        = 2*vdwtype[jnrA+0];
 375             vdwjidx0B        = 2*vdwtype[jnrB+0];
 376             vdwjidx0C        = 2*vdwtype[jnrC+0];
 377             vdwjidx0D        = 2*vdwtype[jnrD+0];
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             r00              = _mm_mul_ps(rsq00,rinv00);
 384             r00              = _mm_andnot_ps(dummy_mask,r00);
 385
 386             /* Compute parameters for interactions between i and j atoms */
 387             qq00             = _mm_mul_ps(iq0,jq0);
 388             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 389                                          vdwparam+vdwioffset0+vdwjidx0B,
 390                                          vdwparam+vdwioffset0+vdwjidx0C,
 391                                          vdwparam+vdwioffset0+vdwjidx0D,
 392                                          &c6_00,&c12_00);
 393
 394             /* Calculate table index by multiplying r with table scale and truncate to integer */
 395             rt               = _mm_mul_ps(r00,vftabscale);
 396             vfitab           = _mm_cvttps_epi32(rt);
 397             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 398             vfitab           = _mm_slli_epi32(vfitab,3);
 399
 400             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 401             isaprod          = _mm_mul_ps(isai0,isaj0);
 402             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 403             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 404
 405             /* Calculate generalized born table index - this is a separate table from the normal one,
 406              * but we use the same procedure by multiplying r with scale and truncating to integer.
 407              */
 408             rt               = _mm_mul_ps(r00,gbscale);
 409             gbitab           = _mm_cvttps_epi32(rt);
 410             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 411             gbitab           = _mm_slli_epi32(gbitab,2);
 412
 413             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 414             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 415             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 416             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 417             _MM_TRANSPOSE4_PS(Y,F,G,H);
 418             Heps             = _mm_mul_ps(gbeps,H);
 419             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 420             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 421             vgb              = _mm_mul_ps(gbqqfactor,VV);
 422
 423             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 424             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 425             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 426             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 427             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 428             /* 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. */
 429             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 430             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 431             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 432             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 433             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 434             velec            = _mm_mul_ps(qq00,rinv00);
 435             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 436
 437             /* CUBIC SPLINE TABLE DISPERSION */
 438             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 439             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 440             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 441             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 442             _MM_TRANSPOSE4_PS(Y,F,G,H);
 443             Heps             = _mm_mul_ps(vfeps,H);
 444             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 445             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 446             vvdw6            = _mm_mul_ps(c6_00,VV);
 447             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 448             fvdw6            = _mm_mul_ps(c6_00,FF);
 449
 450             /* CUBIC SPLINE TABLE REPULSION */
 451             vfitab           = _mm_add_epi32(vfitab,ifour);
 452             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 453             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 454             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 455             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 456             _MM_TRANSPOSE4_PS(Y,F,G,H);
 457             Heps             = _mm_mul_ps(vfeps,H);
 458             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 459             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 460             vvdw12           = _mm_mul_ps(c12_00,VV);
 461             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 462             fvdw12           = _mm_mul_ps(c12_00,FF);
 463             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 464             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 465
 466             /* Update potential sum for this i atom from the interaction with this j atom. */
 467             velec            = _mm_andnot_ps(dummy_mask,velec);
 468             velecsum         = _mm_add_ps(velecsum,velec);
 469             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 470             vgbsum           = _mm_add_ps(vgbsum,vgb);
 471             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 472             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 473
 474             fscal            = _mm_add_ps(felec,fvdw);
 475
 476             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 477
 478             /* Calculate temporary vectorial force */
 479             tx               = _mm_mul_ps(fscal,dx00);
 480             ty               = _mm_mul_ps(fscal,dy00);
 481             tz               = _mm_mul_ps(fscal,dz00);
 482
 483             /* Update vectorial force */
 484             fix0             = _mm_add_ps(fix0,tx);
 485             fiy0             = _mm_add_ps(fiy0,ty);
 486             fiz0             = _mm_add_ps(fiz0,tz);
 487
 488             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 489             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 490             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 491             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 492             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 493
 494             /* Inner loop uses 93 flops */
 495         }
 496
 497         /* End of innermost loop */
 498
 499         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 500                                               f+i_coord_offset,fshift+i_shift_offset);
 501
 502         ggid                        = gid[iidx];
 503         /* Update potential energies */
 504         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 505         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 506         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 507         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 508         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 509
 510         /* Increment number of inner iterations */
 511         inneriter                  += j_index_end - j_index_start;
 512
 513         /* Outer loop uses 10 flops */
 514     }
 515
 516     /* Increment number of outer iterations */
 517     outeriter        += nri;
 518
 519     /* Update outer/inner flops */
 520
 521     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*93);
 522 }
 523 /*
 524  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_single
 525  * Electrostatics interaction: GeneralizedBorn
 526  * VdW interaction:            CubicSplineTable
 527  * Geometry:                   Particle-Particle
 528  * Calculate force/pot:        Force
 529  */
 530 void
 531 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_single
 532                     (t_nblist                    * gmx_restrict       nlist,
 533                      rvec                        * gmx_restrict          xx,
 534                      rvec                        * gmx_restrict          ff,
 535                      t_forcerec                  * gmx_restrict          fr,
 536                      t_mdatoms                   * gmx_restrict     mdatoms,
 537                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 538                      t_nrnb                      * gmx_restrict        nrnb)
 539 {
 540     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 541      * just 0 for non-waters.
 542      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 543      * jnr indices corresponding to data put in the four positions in the SIMD register.
 544      */
 545     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 546     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 547     int              jnrA,jnrB,jnrC,jnrD;
 548     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 549     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 550     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 551     real             rcutoff_scalar;
 552     real             *shiftvec,*fshift,*x,*f;
 553     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 554     real             scratch[4*DIM];
 555     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 556     int              vdwioffset0;
 557     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 558     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 559     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 560     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 561     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 562     real             *charge;
 563     __m128i          gbitab;
 564     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 565     __m128           minushalf = _mm_set1_ps(-0.5);
 566     real             *invsqrta,*dvda,*gbtab;
 567     int              nvdwtype;
 568     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 569     int              *vdwtype;
 570     real             *vdwparam;
 571     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 572     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 573     __m128i          vfitab;
 574     __m128i          ifour       = _mm_set1_epi32(4);
 575     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 576     real             *vftab;
 577     __m128           dummy_mask,cutoff_mask;
 578     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 579     __m128           one     = _mm_set1_ps(1.0);
 580     __m128           two     = _mm_set1_ps(2.0);
 581     x                = xx[0];
 582     f                = ff[0];
 583
 584     nri              = nlist->nri;
 585     iinr             = nlist->iinr;
 586     jindex           = nlist->jindex;
 587     jjnr             = nlist->jjnr;
 588     shiftidx         = nlist->shift;
 589     gid              = nlist->gid;
 590     shiftvec         = fr->shift_vec[0];
 591     fshift           = fr->fshift[0];
 592     facel            = _mm_set1_ps(fr->epsfac);
 593     charge           = mdatoms->chargeA;
 594     nvdwtype         = fr->ntype;
 595     vdwparam         = fr->nbfp;
 596     vdwtype          = mdatoms->typeA;
 597
 598     vftab            = kernel_data->table_vdw->data;
 599     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 600
 601     invsqrta         = fr->invsqrta;
 602     dvda             = fr->dvda;
 603     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 604     gbtab            = fr->gbtab.data;
 605     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 606
 607     /* Avoid stupid compiler warnings */
 608     jnrA = jnrB = jnrC = jnrD = 0;
 609     j_coord_offsetA = 0;
 610     j_coord_offsetB = 0;
 611     j_coord_offsetC = 0;
 612     j_coord_offsetD = 0;
 613
 614     outeriter        = 0;
 615     inneriter        = 0;
 616
 617     for(iidx=0;iidx<4*DIM;iidx++)
 618     {
 619         scratch[iidx] = 0.0;
 620     }
 621
 622     /* Start outer loop over neighborlists */
 623     for(iidx=0; iidx<nri; iidx++)
 624     {
 625         /* Load shift vector for this list */
 626         i_shift_offset   = DIM*shiftidx[iidx];
 627
 628         /* Load limits for loop over neighbors */
 629         j_index_start    = jindex[iidx];
 630         j_index_end      = jindex[iidx+1];
 631
 632         /* Get outer coordinate index */
 633         inr              = iinr[iidx];
 634         i_coord_offset   = DIM*inr;
 635
 636         /* Load i particle coords and add shift vector */
 637         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 638
 639         fix0             = _mm_setzero_ps();
 640         fiy0             = _mm_setzero_ps();
 641         fiz0             = _mm_setzero_ps();
 642
 643         /* Load parameters for i particles */
 644         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 645         isai0            = _mm_load1_ps(invsqrta+inr+0);
 646         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 647
 648         dvdasum          = _mm_setzero_ps();
 649
 650         /* Start inner kernel loop */
 651         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 652         {
 653
 654             /* Get j neighbor index, and coordinate index */
 655             jnrA             = jjnr[jidx];
 656             jnrB             = jjnr[jidx+1];
 657             jnrC             = jjnr[jidx+2];
 658             jnrD             = jjnr[jidx+3];
 659             j_coord_offsetA  = DIM*jnrA;
 660             j_coord_offsetB  = DIM*jnrB;
 661             j_coord_offsetC  = DIM*jnrC;
 662             j_coord_offsetD  = DIM*jnrD;
 663
 664             /* load j atom coordinates */
 665             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 666                                               x+j_coord_offsetC,x+j_coord_offsetD,
 667                                               &jx0,&jy0,&jz0);
 668
 669             /* Calculate displacement vector */
 670             dx00             = _mm_sub_ps(ix0,jx0);
 671             dy00             = _mm_sub_ps(iy0,jy0);
 672             dz00             = _mm_sub_ps(iz0,jz0);
 673
 674             /* Calculate squared distance and things based on it */
 675             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 676
 677             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 678
 679             /* Load parameters for j particles */
 680             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 681                                                               charge+jnrC+0,charge+jnrD+0);
 682             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 683                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 684             vdwjidx0A        = 2*vdwtype[jnrA+0];
 685             vdwjidx0B        = 2*vdwtype[jnrB+0];
 686             vdwjidx0C        = 2*vdwtype[jnrC+0];
 687             vdwjidx0D        = 2*vdwtype[jnrD+0];
 688
 689             /**************************
 690              * CALCULATE INTERACTIONS *
 691              **************************/
 692
 693             r00              = _mm_mul_ps(rsq00,rinv00);
 694
 695             /* Compute parameters for interactions between i and j atoms */
 696             qq00             = _mm_mul_ps(iq0,jq0);
 697             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 698                                          vdwparam+vdwioffset0+vdwjidx0B,
 699                                          vdwparam+vdwioffset0+vdwjidx0C,
 700                                          vdwparam+vdwioffset0+vdwjidx0D,
 701                                          &c6_00,&c12_00);
 702
 703             /* Calculate table index by multiplying r with table scale and truncate to integer */
 704             rt               = _mm_mul_ps(r00,vftabscale);
 705             vfitab           = _mm_cvttps_epi32(rt);
 706             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 707             vfitab           = _mm_slli_epi32(vfitab,3);
 708
 709             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 710             isaprod          = _mm_mul_ps(isai0,isaj0);
 711             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 712             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 713
 714             /* Calculate generalized born table index - this is a separate table from the normal one,
 715              * but we use the same procedure by multiplying r with scale and truncating to integer.
 716              */
 717             rt               = _mm_mul_ps(r00,gbscale);
 718             gbitab           = _mm_cvttps_epi32(rt);
 719             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 720             gbitab           = _mm_slli_epi32(gbitab,2);
 721
 722             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 723             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 724             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 725             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 726             _MM_TRANSPOSE4_PS(Y,F,G,H);
 727             Heps             = _mm_mul_ps(gbeps,H);
 728             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 729             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 730             vgb              = _mm_mul_ps(gbqqfactor,VV);
 731
 732             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 733             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 734             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 735             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 736             fjptrA           = dvda+jnrA;
 737             fjptrB           = dvda+jnrB;
 738             fjptrC           = dvda+jnrC;
 739             fjptrD           = dvda+jnrD;
 740             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 741             velec            = _mm_mul_ps(qq00,rinv00);
 742             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 743
 744             /* CUBIC SPLINE TABLE DISPERSION */
 745             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 746             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 747             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 748             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 749             _MM_TRANSPOSE4_PS(Y,F,G,H);
 750             Heps             = _mm_mul_ps(vfeps,H);
 751             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 752             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 753             fvdw6            = _mm_mul_ps(c6_00,FF);
 754
 755             /* CUBIC SPLINE TABLE REPULSION */
 756             vfitab           = _mm_add_epi32(vfitab,ifour);
 757             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 758             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 759             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 760             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 761             _MM_TRANSPOSE4_PS(Y,F,G,H);
 762             Heps             = _mm_mul_ps(vfeps,H);
 763             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 764             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 765             fvdw12           = _mm_mul_ps(c12_00,FF);
 766             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 767
 768             fscal            = _mm_add_ps(felec,fvdw);
 769
 770             /* Calculate temporary vectorial force */
 771             tx               = _mm_mul_ps(fscal,dx00);
 772             ty               = _mm_mul_ps(fscal,dy00);
 773             tz               = _mm_mul_ps(fscal,dz00);
 774
 775             /* Update vectorial force */
 776             fix0             = _mm_add_ps(fix0,tx);
 777             fiy0             = _mm_add_ps(fiy0,ty);
 778             fiz0             = _mm_add_ps(fiz0,tz);
 779
 780             fjptrA             = f+j_coord_offsetA;
 781             fjptrB             = f+j_coord_offsetB;
 782             fjptrC             = f+j_coord_offsetC;
 783             fjptrD             = f+j_coord_offsetD;
 784             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 785
 786             /* Inner loop uses 82 flops */
 787         }
 788
 789         if(jidx<j_index_end)
 790         {
 791
 792             /* Get j neighbor index, and coordinate index */
 793             jnrlistA         = jjnr[jidx];
 794             jnrlistB         = jjnr[jidx+1];
 795             jnrlistC         = jjnr[jidx+2];
 796             jnrlistD         = jjnr[jidx+3];
 797             /* Sign of each element will be negative for non-real atoms.
 798              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 799              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 800              */
 801             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 802             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 803             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 804             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 805             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 806             j_coord_offsetA  = DIM*jnrA;
 807             j_coord_offsetB  = DIM*jnrB;
 808             j_coord_offsetC  = DIM*jnrC;
 809             j_coord_offsetD  = DIM*jnrD;
 810
 811             /* load j atom coordinates */
 812             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 813                                               x+j_coord_offsetC,x+j_coord_offsetD,
 814                                               &jx0,&jy0,&jz0);
 815
 816             /* Calculate displacement vector */
 817             dx00             = _mm_sub_ps(ix0,jx0);
 818             dy00             = _mm_sub_ps(iy0,jy0);
 819             dz00             = _mm_sub_ps(iz0,jz0);
 820
 821             /* Calculate squared distance and things based on it */
 822             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 823
 824             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 825
 826             /* Load parameters for j particles */
 827             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 828                                                               charge+jnrC+0,charge+jnrD+0);
 829             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 830                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 831             vdwjidx0A        = 2*vdwtype[jnrA+0];
 832             vdwjidx0B        = 2*vdwtype[jnrB+0];
 833             vdwjidx0C        = 2*vdwtype[jnrC+0];
 834             vdwjidx0D        = 2*vdwtype[jnrD+0];
 835
 836             /**************************
 837              * CALCULATE INTERACTIONS *
 838              **************************/
 839
 840             r00              = _mm_mul_ps(rsq00,rinv00);
 841             r00              = _mm_andnot_ps(dummy_mask,r00);
 842
 843             /* Compute parameters for interactions between i and j atoms */
 844             qq00             = _mm_mul_ps(iq0,jq0);
 845             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 846                                          vdwparam+vdwioffset0+vdwjidx0B,
 847                                          vdwparam+vdwioffset0+vdwjidx0C,
 848                                          vdwparam+vdwioffset0+vdwjidx0D,
 849                                          &c6_00,&c12_00);
 850
 851             /* Calculate table index by multiplying r with table scale and truncate to integer */
 852             rt               = _mm_mul_ps(r00,vftabscale);
 853             vfitab           = _mm_cvttps_epi32(rt);
 854             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 855             vfitab           = _mm_slli_epi32(vfitab,3);
 856
 857             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 858             isaprod          = _mm_mul_ps(isai0,isaj0);
 859             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 860             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 861
 862             /* Calculate generalized born table index - this is a separate table from the normal one,
 863              * but we use the same procedure by multiplying r with scale and truncating to integer.
 864              */
 865             rt               = _mm_mul_ps(r00,gbscale);
 866             gbitab           = _mm_cvttps_epi32(rt);
 867             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 868             gbitab           = _mm_slli_epi32(gbitab,2);
 869
 870             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 871             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 872             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 873             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 874             _MM_TRANSPOSE4_PS(Y,F,G,H);
 875             Heps             = _mm_mul_ps(gbeps,H);
 876             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 877             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 878             vgb              = _mm_mul_ps(gbqqfactor,VV);
 879
 880             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 881             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 882             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 883             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 884             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 885             /* 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. */
 886             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 887             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 888             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 889             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 890             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 891             velec            = _mm_mul_ps(qq00,rinv00);
 892             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 893
 894             /* CUBIC SPLINE TABLE DISPERSION */
 895             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 896             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 897             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 898             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 899             _MM_TRANSPOSE4_PS(Y,F,G,H);
 900             Heps             = _mm_mul_ps(vfeps,H);
 901             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 902             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 903             fvdw6            = _mm_mul_ps(c6_00,FF);
 904
 905             /* CUBIC SPLINE TABLE REPULSION */
 906             vfitab           = _mm_add_epi32(vfitab,ifour);
 907             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 908             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 909             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 910             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 911             _MM_TRANSPOSE4_PS(Y,F,G,H);
 912             Heps             = _mm_mul_ps(vfeps,H);
 913             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 914             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 915             fvdw12           = _mm_mul_ps(c12_00,FF);
 916             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 917
 918             fscal            = _mm_add_ps(felec,fvdw);
 919
 920             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm_mul_ps(fscal,dx00);
 924             ty               = _mm_mul_ps(fscal,dy00);
 925             tz               = _mm_mul_ps(fscal,dz00);
 926
 927             /* Update vectorial force */
 928             fix0             = _mm_add_ps(fix0,tx);
 929             fiy0             = _mm_add_ps(fiy0,ty);
 930             fiz0             = _mm_add_ps(fiz0,tz);
 931
 932             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 933             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 934             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 935             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 936             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 937
 938             /* Inner loop uses 83 flops */
 939         }
 940
 941         /* End of innermost loop */
 942
 943         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 944                                               f+i_coord_offset,fshift+i_shift_offset);
 945
 946         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 947         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 948
 949         /* Increment number of inner iterations */
 950         inneriter                  += j_index_end - j_index_start;
 951
 952         /* Outer loop uses 7 flops */
 953     }
 954
 955     /* Increment number of outer iterations */
 956     outeriter        += nri;
 957
 958     /* Update outer/inner flops */
 959
 960     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);
 961 }