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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: GeneralizedBorn
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     __m128i          gbitab;
  91     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  92     __m128           minushalf = _mm_set1_ps(-0.5);
  93     real             *invsqrta,*dvda,*gbtab;
  94     int              nvdwtype;
  95     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  96     int              *vdwtype;
  97     real             *vdwparam;
  98     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  99     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 100     __m128i          vfitab;
 101     __m128i          ifour       = _mm_set1_epi32(4);
 102     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 103     real             *vftab;
 104     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     vftab            = kernel_data->table_vdw->data;
 126     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 127
 128     invsqrta         = fr->invsqrta;
 129     dvda             = fr->dvda;
 130     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 131     gbtab            = fr->gbtab.data;
 132     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = jnrC = jnrD = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138     j_coord_offsetC = 0;
 139     j_coord_offsetD = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 165
 166         fix0             = _mm_setzero_ps();
 167         fiy0             = _mm_setzero_ps();
 168         fiz0             = _mm_setzero_ps();
 169
 170         /* Load parameters for i particles */
 171         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 172         isai0            = _mm_load1_ps(invsqrta+inr+0);
 173         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_ps();
 177         vgbsum           = _mm_setzero_ps();
 178         vvdwsum          = _mm_setzero_ps();
 179         dvdasum          = _mm_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192             j_coord_offsetC  = DIM*jnrC;
 193             j_coord_offsetD  = DIM*jnrD;
 194
 195             /* load j atom coordinates */
 196             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                               x+j_coord_offsetC,x+j_coord_offsetD,
 198                                               &jx0,&jy0,&jz0);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm_sub_ps(ix0,jx0);
 202             dy00             = _mm_sub_ps(iy0,jy0);
 203             dz00             = _mm_sub_ps(iz0,jz0);
 204
 205             /* Calculate squared distance and things based on it */
 206             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 207
 208             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 209
 210             /* Load parameters for j particles */
 211             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 212                                                               charge+jnrC+0,charge+jnrD+0);
 213             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 214                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 215             vdwjidx0A        = 2*vdwtype[jnrA+0];
 216             vdwjidx0B        = 2*vdwtype[jnrB+0];
 217             vdwjidx0C        = 2*vdwtype[jnrC+0];
 218             vdwjidx0D        = 2*vdwtype[jnrD+0];
 219
 220             /**************************
 221              * CALCULATE INTERACTIONS *
 222              **************************/
 223
 224             r00              = _mm_mul_ps(rsq00,rinv00);
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq00             = _mm_mul_ps(iq0,jq0);
 228             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 229                                          vdwparam+vdwioffset0+vdwjidx0B,
 230                                          vdwparam+vdwioffset0+vdwjidx0C,
 231                                          vdwparam+vdwioffset0+vdwjidx0D,
 232                                          &c6_00,&c12_00);
 233
 234             /* Calculate table index by multiplying r with table scale and truncate to integer */
 235             rt               = _mm_mul_ps(r00,vftabscale);
 236             vfitab           = _mm_cvttps_epi32(rt);
 237             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 238             vfitab           = _mm_slli_epi32(vfitab,3);
 239
 240             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 241             isaprod          = _mm_mul_ps(isai0,isaj0);
 242             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 243             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 244
 245             /* Calculate generalized born table index - this is a separate table from the normal one,
 246              * but we use the same procedure by multiplying r with scale and truncating to integer.
 247              */
 248             rt               = _mm_mul_ps(r00,gbscale);
 249             gbitab           = _mm_cvttps_epi32(rt);
 250             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 251             gbitab           = _mm_slli_epi32(gbitab,2);
 252
 253             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 254             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 255             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 256             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 257             _MM_TRANSPOSE4_PS(Y,F,G,H);
 258             Heps             = _mm_mul_ps(gbeps,H);
 259             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 260             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 261             vgb              = _mm_mul_ps(gbqqfactor,VV);
 262
 263             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 264             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 265             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 266             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 267             fjptrA           = dvda+jnrA;
 268             fjptrB           = dvda+jnrB;
 269             fjptrC           = dvda+jnrC;
 270             fjptrD           = dvda+jnrD;
 271             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 272             velec            = _mm_mul_ps(qq00,rinv00);
 273             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 274
 275             /* CUBIC SPLINE TABLE DISPERSION */
 276             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 277             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 278             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 279             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 280             _MM_TRANSPOSE4_PS(Y,F,G,H);
 281             Heps             = _mm_mul_ps(vfeps,H);
 282             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 283             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 284             vvdw6            = _mm_mul_ps(c6_00,VV);
 285             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 286             fvdw6            = _mm_mul_ps(c6_00,FF);
 287
 288             /* CUBIC SPLINE TABLE REPULSION */
 289             vfitab           = _mm_add_epi32(vfitab,ifour);
 290             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 291             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 292             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 293             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 294             _MM_TRANSPOSE4_PS(Y,F,G,H);
 295             Heps             = _mm_mul_ps(vfeps,H);
 296             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 297             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 298             vvdw12           = _mm_mul_ps(c12_00,VV);
 299             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 300             fvdw12           = _mm_mul_ps(c12_00,FF);
 301             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 302             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velecsum         = _mm_add_ps(velecsum,velec);
 306             vgbsum           = _mm_add_ps(vgbsum,vgb);
 307             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 308
 309             fscal            = _mm_add_ps(felec,fvdw);
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm_mul_ps(fscal,dx00);
 313             ty               = _mm_mul_ps(fscal,dy00);
 314             tz               = _mm_mul_ps(fscal,dz00);
 315
 316             /* Update vectorial force */
 317             fix0             = _mm_add_ps(fix0,tx);
 318             fiy0             = _mm_add_ps(fiy0,ty);
 319             fiz0             = _mm_add_ps(fiz0,tz);
 320
 321             fjptrA             = f+j_coord_offsetA;
 322             fjptrB             = f+j_coord_offsetB;
 323             fjptrC             = f+j_coord_offsetC;
 324             fjptrD             = f+j_coord_offsetD;
 325             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 326
 327             /* Inner loop uses 92 flops */
 328         }
 329
 330         if(jidx<j_index_end)
 331         {
 332
 333             /* Get j neighbor index, and coordinate index */
 334             jnrlistA         = jjnr[jidx];
 335             jnrlistB         = jjnr[jidx+1];
 336             jnrlistC         = jjnr[jidx+2];
 337             jnrlistD         = jjnr[jidx+3];
 338             /* Sign of each element will be negative for non-real atoms.
 339              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 340              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 341              */
 342             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 343             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 344             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 345             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 346             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 347             j_coord_offsetA  = DIM*jnrA;
 348             j_coord_offsetB  = DIM*jnrB;
 349             j_coord_offsetC  = DIM*jnrC;
 350             j_coord_offsetD  = DIM*jnrD;
 351
 352             /* load j atom coordinates */
 353             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 354                                               x+j_coord_offsetC,x+j_coord_offsetD,
 355                                               &jx0,&jy0,&jz0);
 356
 357             /* Calculate displacement vector */
 358             dx00             = _mm_sub_ps(ix0,jx0);
 359             dy00             = _mm_sub_ps(iy0,jy0);
 360             dz00             = _mm_sub_ps(iz0,jz0);
 361
 362             /* Calculate squared distance and things based on it */
 363             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 364
 365             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 366
 367             /* Load parameters for j particles */
 368             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 369                                                               charge+jnrC+0,charge+jnrD+0);
 370             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 371                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 372             vdwjidx0A        = 2*vdwtype[jnrA+0];
 373             vdwjidx0B        = 2*vdwtype[jnrB+0];
 374             vdwjidx0C        = 2*vdwtype[jnrC+0];
 375             vdwjidx0D        = 2*vdwtype[jnrD+0];
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             r00              = _mm_mul_ps(rsq00,rinv00);
 382             r00              = _mm_andnot_ps(dummy_mask,r00);
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq00             = _mm_mul_ps(iq0,jq0);
 386             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 387                                          vdwparam+vdwioffset0+vdwjidx0B,
 388                                          vdwparam+vdwioffset0+vdwjidx0C,
 389                                          vdwparam+vdwioffset0+vdwjidx0D,
 390                                          &c6_00,&c12_00);
 391
 392             /* Calculate table index by multiplying r with table scale and truncate to integer */
 393             rt               = _mm_mul_ps(r00,vftabscale);
 394             vfitab           = _mm_cvttps_epi32(rt);
 395             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 396             vfitab           = _mm_slli_epi32(vfitab,3);
 397
 398             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 399             isaprod          = _mm_mul_ps(isai0,isaj0);
 400             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 401             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 402
 403             /* Calculate generalized born table index - this is a separate table from the normal one,
 404              * but we use the same procedure by multiplying r with scale and truncating to integer.
 405              */
 406             rt               = _mm_mul_ps(r00,gbscale);
 407             gbitab           = _mm_cvttps_epi32(rt);
 408             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 409             gbitab           = _mm_slli_epi32(gbitab,2);
 410
 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             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 425             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 426             /* 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. */
 427             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 428             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 429             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 430             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 431             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 432             velec            = _mm_mul_ps(qq00,rinv00);
 433             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 434
 435             /* CUBIC SPLINE TABLE DISPERSION */
 436             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 437             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 438             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 439             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 440             _MM_TRANSPOSE4_PS(Y,F,G,H);
 441             Heps             = _mm_mul_ps(vfeps,H);
 442             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 443             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 444             vvdw6            = _mm_mul_ps(c6_00,VV);
 445             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 446             fvdw6            = _mm_mul_ps(c6_00,FF);
 447
 448             /* CUBIC SPLINE TABLE REPULSION */
 449             vfitab           = _mm_add_epi32(vfitab,ifour);
 450             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 451             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 452             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 453             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 454             _MM_TRANSPOSE4_PS(Y,F,G,H);
 455             Heps             = _mm_mul_ps(vfeps,H);
 456             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 457             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 458             vvdw12           = _mm_mul_ps(c12_00,VV);
 459             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 460             fvdw12           = _mm_mul_ps(c12_00,FF);
 461             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 462             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 463
 464             /* Update potential sum for this i atom from the interaction with this j atom. */
 465             velec            = _mm_andnot_ps(dummy_mask,velec);
 466             velecsum         = _mm_add_ps(velecsum,velec);
 467             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 468             vgbsum           = _mm_add_ps(vgbsum,vgb);
 469             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 470             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 471
 472             fscal            = _mm_add_ps(felec,fvdw);
 473
 474             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 475
 476             /* Calculate temporary vectorial force */
 477             tx               = _mm_mul_ps(fscal,dx00);
 478             ty               = _mm_mul_ps(fscal,dy00);
 479             tz               = _mm_mul_ps(fscal,dz00);
 480
 481             /* Update vectorial force */
 482             fix0             = _mm_add_ps(fix0,tx);
 483             fiy0             = _mm_add_ps(fiy0,ty);
 484             fiz0             = _mm_add_ps(fiz0,tz);
 485
 486             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 487             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 488             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 489             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 490             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 491
 492             /* Inner loop uses 93 flops */
 493         }
 494
 495         /* End of innermost loop */
 496
 497         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 498                                               f+i_coord_offset,fshift+i_shift_offset);
 499
 500         ggid                        = gid[iidx];
 501         /* Update potential energies */
 502         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 503         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 504         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 505         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 506         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 507
 508         /* Increment number of inner iterations */
 509         inneriter                  += j_index_end - j_index_start;
 510
 511         /* Outer loop uses 10 flops */
 512     }
 513
 514     /* Increment number of outer iterations */
 515     outeriter        += nri;
 516
 517     /* Update outer/inner flops */
 518
 519     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*93);
 520 }
 521 /*
 522  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_single
 523  * Electrostatics interaction: GeneralizedBorn
 524  * VdW interaction:            CubicSplineTable
 525  * Geometry:                   Particle-Particle
 526  * Calculate force/pot:        Force
 527  */
 528 void
 529 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_single
 530                     (t_nblist                    * gmx_restrict       nlist,
 531                      rvec                        * gmx_restrict          xx,
 532                      rvec                        * gmx_restrict          ff,
 533                      t_forcerec                  * gmx_restrict          fr,
 534                      t_mdatoms                   * gmx_restrict     mdatoms,
 535                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 536                      t_nrnb                      * gmx_restrict        nrnb)
 537 {
 538     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 539      * just 0 for non-waters.
 540      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 541      * jnr indices corresponding to data put in the four positions in the SIMD register.
 542      */
 543     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 544     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 545     int              jnrA,jnrB,jnrC,jnrD;
 546     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 547     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 548     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 549     real             rcutoff_scalar;
 550     real             *shiftvec,*fshift,*x,*f;
 551     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 552     real             scratch[4*DIM];
 553     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 554     int              vdwioffset0;
 555     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 556     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 557     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 558     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 559     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 560     real             *charge;
 561     __m128i          gbitab;
 562     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 563     __m128           minushalf = _mm_set1_ps(-0.5);
 564     real             *invsqrta,*dvda,*gbtab;
 565     int              nvdwtype;
 566     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 567     int              *vdwtype;
 568     real             *vdwparam;
 569     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 570     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 571     __m128i          vfitab;
 572     __m128i          ifour       = _mm_set1_epi32(4);
 573     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 574     real             *vftab;
 575     __m128           dummy_mask,cutoff_mask;
 576     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 577     __m128           one     = _mm_set1_ps(1.0);
 578     __m128           two     = _mm_set1_ps(2.0);
 579     x                = xx[0];
 580     f                = ff[0];
 581
 582     nri              = nlist->nri;
 583     iinr             = nlist->iinr;
 584     jindex           = nlist->jindex;
 585     jjnr             = nlist->jjnr;
 586     shiftidx         = nlist->shift;
 587     gid              = nlist->gid;
 588     shiftvec         = fr->shift_vec[0];
 589     fshift           = fr->fshift[0];
 590     facel            = _mm_set1_ps(fr->epsfac);
 591     charge           = mdatoms->chargeA;
 592     nvdwtype         = fr->ntype;
 593     vdwparam         = fr->nbfp;
 594     vdwtype          = mdatoms->typeA;
 595
 596     vftab            = kernel_data->table_vdw->data;
 597     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 598
 599     invsqrta         = fr->invsqrta;
 600     dvda             = fr->dvda;
 601     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 602     gbtab            = fr->gbtab.data;
 603     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 604
 605     /* Avoid stupid compiler warnings */
 606     jnrA = jnrB = jnrC = jnrD = 0;
 607     j_coord_offsetA = 0;
 608     j_coord_offsetB = 0;
 609     j_coord_offsetC = 0;
 610     j_coord_offsetD = 0;
 611
 612     outeriter        = 0;
 613     inneriter        = 0;
 614
 615     for(iidx=0;iidx<4*DIM;iidx++)
 616     {
 617         scratch[iidx] = 0.0;
 618     }
 619
 620     /* Start outer loop over neighborlists */
 621     for(iidx=0; iidx<nri; iidx++)
 622     {
 623         /* Load shift vector for this list */
 624         i_shift_offset   = DIM*shiftidx[iidx];
 625
 626         /* Load limits for loop over neighbors */
 627         j_index_start    = jindex[iidx];
 628         j_index_end      = jindex[iidx+1];
 629
 630         /* Get outer coordinate index */
 631         inr              = iinr[iidx];
 632         i_coord_offset   = DIM*inr;
 633
 634         /* Load i particle coords and add shift vector */
 635         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 636
 637         fix0             = _mm_setzero_ps();
 638         fiy0             = _mm_setzero_ps();
 639         fiz0             = _mm_setzero_ps();
 640
 641         /* Load parameters for i particles */
 642         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 643         isai0            = _mm_load1_ps(invsqrta+inr+0);
 644         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 645
 646         dvdasum          = _mm_setzero_ps();
 647
 648         /* Start inner kernel loop */
 649         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 650         {
 651
 652             /* Get j neighbor index, and coordinate index */
 653             jnrA             = jjnr[jidx];
 654             jnrB             = jjnr[jidx+1];
 655             jnrC             = jjnr[jidx+2];
 656             jnrD             = jjnr[jidx+3];
 657             j_coord_offsetA  = DIM*jnrA;
 658             j_coord_offsetB  = DIM*jnrB;
 659             j_coord_offsetC  = DIM*jnrC;
 660             j_coord_offsetD  = DIM*jnrD;
 661
 662             /* load j atom coordinates */
 663             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 664                                               x+j_coord_offsetC,x+j_coord_offsetD,
 665                                               &jx0,&jy0,&jz0);
 666
 667             /* Calculate displacement vector */
 668             dx00             = _mm_sub_ps(ix0,jx0);
 669             dy00             = _mm_sub_ps(iy0,jy0);
 670             dz00             = _mm_sub_ps(iz0,jz0);
 671
 672             /* Calculate squared distance and things based on it */
 673             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 674
 675             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 676
 677             /* Load parameters for j particles */
 678             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 679                                                               charge+jnrC+0,charge+jnrD+0);
 680             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 681                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 682             vdwjidx0A        = 2*vdwtype[jnrA+0];
 683             vdwjidx0B        = 2*vdwtype[jnrB+0];
 684             vdwjidx0C        = 2*vdwtype[jnrC+0];
 685             vdwjidx0D        = 2*vdwtype[jnrD+0];
 686
 687             /**************************
 688              * CALCULATE INTERACTIONS *
 689              **************************/
 690
 691             r00              = _mm_mul_ps(rsq00,rinv00);
 692
 693             /* Compute parameters for interactions between i and j atoms */
 694             qq00             = _mm_mul_ps(iq0,jq0);
 695             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 696                                          vdwparam+vdwioffset0+vdwjidx0B,
 697                                          vdwparam+vdwioffset0+vdwjidx0C,
 698                                          vdwparam+vdwioffset0+vdwjidx0D,
 699                                          &c6_00,&c12_00);
 700
 701             /* Calculate table index by multiplying r with table scale and truncate to integer */
 702             rt               = _mm_mul_ps(r00,vftabscale);
 703             vfitab           = _mm_cvttps_epi32(rt);
 704             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 705             vfitab           = _mm_slli_epi32(vfitab,3);
 706
 707             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 708             isaprod          = _mm_mul_ps(isai0,isaj0);
 709             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 710             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 711
 712             /* Calculate generalized born table index - this is a separate table from the normal one,
 713              * but we use the same procedure by multiplying r with scale and truncating to integer.
 714              */
 715             rt               = _mm_mul_ps(r00,gbscale);
 716             gbitab           = _mm_cvttps_epi32(rt);
 717             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 718             gbitab           = _mm_slli_epi32(gbitab,2);
 719
 720             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 721             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 722             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 723             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 724             _MM_TRANSPOSE4_PS(Y,F,G,H);
 725             Heps             = _mm_mul_ps(gbeps,H);
 726             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 727             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 728             vgb              = _mm_mul_ps(gbqqfactor,VV);
 729
 730             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 731             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 732             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 733             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 734             fjptrA           = dvda+jnrA;
 735             fjptrB           = dvda+jnrB;
 736             fjptrC           = dvda+jnrC;
 737             fjptrD           = dvda+jnrD;
 738             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 739             velec            = _mm_mul_ps(qq00,rinv00);
 740             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 741
 742             /* CUBIC SPLINE TABLE DISPERSION */
 743             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 744             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 745             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 746             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 747             _MM_TRANSPOSE4_PS(Y,F,G,H);
 748             Heps             = _mm_mul_ps(vfeps,H);
 749             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 750             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 751             fvdw6            = _mm_mul_ps(c6_00,FF);
 752
 753             /* CUBIC SPLINE TABLE REPULSION */
 754             vfitab           = _mm_add_epi32(vfitab,ifour);
 755             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 756             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 757             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 758             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 759             _MM_TRANSPOSE4_PS(Y,F,G,H);
 760             Heps             = _mm_mul_ps(vfeps,H);
 761             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 762             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 763             fvdw12           = _mm_mul_ps(c12_00,FF);
 764             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 765
 766             fscal            = _mm_add_ps(felec,fvdw);
 767
 768             /* Calculate temporary vectorial force */
 769             tx               = _mm_mul_ps(fscal,dx00);
 770             ty               = _mm_mul_ps(fscal,dy00);
 771             tz               = _mm_mul_ps(fscal,dz00);
 772
 773             /* Update vectorial force */
 774             fix0             = _mm_add_ps(fix0,tx);
 775             fiy0             = _mm_add_ps(fiy0,ty);
 776             fiz0             = _mm_add_ps(fiz0,tz);
 777
 778             fjptrA             = f+j_coord_offsetA;
 779             fjptrB             = f+j_coord_offsetB;
 780             fjptrC             = f+j_coord_offsetC;
 781             fjptrD             = f+j_coord_offsetD;
 782             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 783
 784             /* Inner loop uses 82 flops */
 785         }
 786
 787         if(jidx<j_index_end)
 788         {
 789
 790             /* Get j neighbor index, and coordinate index */
 791             jnrlistA         = jjnr[jidx];
 792             jnrlistB         = jjnr[jidx+1];
 793             jnrlistC         = jjnr[jidx+2];
 794             jnrlistD         = jjnr[jidx+3];
 795             /* Sign of each element will be negative for non-real atoms.
 796              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 797              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 798              */
 799             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 800             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 801             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 802             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 803             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 804             j_coord_offsetA  = DIM*jnrA;
 805             j_coord_offsetB  = DIM*jnrB;
 806             j_coord_offsetC  = DIM*jnrC;
 807             j_coord_offsetD  = DIM*jnrD;
 808
 809             /* load j atom coordinates */
 810             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 811                                               x+j_coord_offsetC,x+j_coord_offsetD,
 812                                               &jx0,&jy0,&jz0);
 813
 814             /* Calculate displacement vector */
 815             dx00             = _mm_sub_ps(ix0,jx0);
 816             dy00             = _mm_sub_ps(iy0,jy0);
 817             dz00             = _mm_sub_ps(iz0,jz0);
 818
 819             /* Calculate squared distance and things based on it */
 820             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 821
 822             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 823
 824             /* Load parameters for j particles */
 825             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 826                                                               charge+jnrC+0,charge+jnrD+0);
 827             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 828                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 829             vdwjidx0A        = 2*vdwtype[jnrA+0];
 830             vdwjidx0B        = 2*vdwtype[jnrB+0];
 831             vdwjidx0C        = 2*vdwtype[jnrC+0];
 832             vdwjidx0D        = 2*vdwtype[jnrD+0];
 833
 834             /**************************
 835              * CALCULATE INTERACTIONS *
 836              **************************/
 837
 838             r00              = _mm_mul_ps(rsq00,rinv00);
 839             r00              = _mm_andnot_ps(dummy_mask,r00);
 840
 841             /* Compute parameters for interactions between i and j atoms */
 842             qq00             = _mm_mul_ps(iq0,jq0);
 843             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 844                                          vdwparam+vdwioffset0+vdwjidx0B,
 845                                          vdwparam+vdwioffset0+vdwjidx0C,
 846                                          vdwparam+vdwioffset0+vdwjidx0D,
 847                                          &c6_00,&c12_00);
 848
 849             /* Calculate table index by multiplying r with table scale and truncate to integer */
 850             rt               = _mm_mul_ps(r00,vftabscale);
 851             vfitab           = _mm_cvttps_epi32(rt);
 852             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 853             vfitab           = _mm_slli_epi32(vfitab,3);
 854
 855             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 856             isaprod          = _mm_mul_ps(isai0,isaj0);
 857             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 858             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 859
 860             /* Calculate generalized born table index - this is a separate table from the normal one,
 861              * but we use the same procedure by multiplying r with scale and truncating to integer.
 862              */
 863             rt               = _mm_mul_ps(r00,gbscale);
 864             gbitab           = _mm_cvttps_epi32(rt);
 865             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 866             gbitab           = _mm_slli_epi32(gbitab,2);
 867
 868             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 869             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 870             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 871             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 872             _MM_TRANSPOSE4_PS(Y,F,G,H);
 873             Heps             = _mm_mul_ps(gbeps,H);
 874             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 875             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 876             vgb              = _mm_mul_ps(gbqqfactor,VV);
 877
 878             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 879             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 880             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 881             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 882             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 883             /* 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. */
 884             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 885             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 886             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 887             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 888             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 889             velec            = _mm_mul_ps(qq00,rinv00);
 890             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 891
 892             /* CUBIC SPLINE TABLE DISPERSION */
 893             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 894             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 895             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 896             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 897             _MM_TRANSPOSE4_PS(Y,F,G,H);
 898             Heps             = _mm_mul_ps(vfeps,H);
 899             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 900             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 901             fvdw6            = _mm_mul_ps(c6_00,FF);
 902
 903             /* CUBIC SPLINE TABLE REPULSION */
 904             vfitab           = _mm_add_epi32(vfitab,ifour);
 905             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 906             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 907             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 908             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 909             _MM_TRANSPOSE4_PS(Y,F,G,H);
 910             Heps             = _mm_mul_ps(vfeps,H);
 911             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 912             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 913             fvdw12           = _mm_mul_ps(c12_00,FF);
 914             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 915
 916             fscal            = _mm_add_ps(felec,fvdw);
 917
 918             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 919
 920             /* Calculate temporary vectorial force */
 921             tx               = _mm_mul_ps(fscal,dx00);
 922             ty               = _mm_mul_ps(fscal,dy00);
 923             tz               = _mm_mul_ps(fscal,dz00);
 924
 925             /* Update vectorial force */
 926             fix0             = _mm_add_ps(fix0,tx);
 927             fiy0             = _mm_add_ps(fiy0,ty);
 928             fiz0             = _mm_add_ps(fiz0,tz);
 929
 930             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 931             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 932             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 933             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 934             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 935
 936             /* Inner loop uses 83 flops */
 937         }
 938
 939         /* End of innermost loop */
 940
 941         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 942                                               f+i_coord_offset,fshift+i_shift_offset);
 943
 944         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 945         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 946
 947         /* Increment number of inner iterations */
 948         inneriter                  += j_index_end - j_index_start;
 949
 950         /* Outer loop uses 7 flops */
 951     }
 952
 953     /* Increment number of outer iterations */
 954     outeriter        += nri;
 955
 956     /* Update outer/inner flops */
 957
 958     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);
 959 }