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

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: GeneralizedBorn
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  75     real             *charge;
  76     __m128i          gbitab;
  77     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  78     __m128           minushalf = _mm_set1_ps(-0.5);
  79     real             *invsqrta,*dvda,*gbtab;
  80     int              nvdwtype;
  81     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  82     int              *vdwtype;
  83     real             *vdwparam;
  84     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  85     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  86     __m128i          vfitab;
  87     __m128i          ifour       = _mm_set1_epi32(4);
  88     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  89     real             *vftab;
  90     __m128           dummy_mask,cutoff_mask;
  91     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  92     __m128           one     = _mm_set1_ps(1.0);
  93     __m128           two     = _mm_set1_ps(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm_set1_ps(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107     nvdwtype         = fr->ntype;
 108     vdwparam         = fr->nbfp;
 109     vdwtype          = mdatoms->typeA;
 110
 111     vftab            = kernel_data->table_vdw->data;
 112     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 113
 114     invsqrta         = fr->invsqrta;
 115     dvda             = fr->dvda;
 116     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 117     gbtab            = fr->gbtab.data;
 118     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = jnrC = jnrD = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124     j_coord_offsetC = 0;
 125     j_coord_offsetD = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     for(iidx=0;iidx<4*DIM;iidx++)
 131     {
 132         scratch[iidx] = 0.0;
 133     }
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _mm_setzero_ps();
 153         fiy0             = _mm_setzero_ps();
 154         fiz0             = _mm_setzero_ps();
 155
 156         /* Load parameters for i particles */
 157         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 158         isai0            = _mm_load1_ps(invsqrta+inr+0);
 159         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 160
 161         /* Reset potential sums */
 162         velecsum         = _mm_setzero_ps();
 163         vgbsum           = _mm_setzero_ps();
 164         vvdwsum          = _mm_setzero_ps();
 165         dvdasum          = _mm_setzero_ps();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             jnrC             = jjnr[jidx+2];
 175             jnrD             = jjnr[jidx+3];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178             j_coord_offsetC  = DIM*jnrC;
 179             j_coord_offsetD  = DIM*jnrD;
 180
 181             /* load j atom coordinates */
 182             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 183                                               x+j_coord_offsetC,x+j_coord_offsetD,
 184                                               &jx0,&jy0,&jz0);
 185
 186             /* Calculate displacement vector */
 187             dx00             = _mm_sub_ps(ix0,jx0);
 188             dy00             = _mm_sub_ps(iy0,jy0);
 189             dz00             = _mm_sub_ps(iz0,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 193
 194             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 198                                                               charge+jnrC+0,charge+jnrD+0);
 199             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 200                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 201             vdwjidx0A        = 2*vdwtype[jnrA+0];
 202             vdwjidx0B        = 2*vdwtype[jnrB+0];
 203             vdwjidx0C        = 2*vdwtype[jnrC+0];
 204             vdwjidx0D        = 2*vdwtype[jnrD+0];
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             r00              = _mm_mul_ps(rsq00,rinv00);
 211
 212             /* Compute parameters for interactions between i and j atoms */
 213             qq00             = _mm_mul_ps(iq0,jq0);
 214             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 215                                          vdwparam+vdwioffset0+vdwjidx0B,
 216                                          vdwparam+vdwioffset0+vdwjidx0C,
 217                                          vdwparam+vdwioffset0+vdwjidx0D,
 218                                          &c6_00,&c12_00);
 219
 220             /* Calculate table index by multiplying r with table scale and truncate to integer */
 221             rt               = _mm_mul_ps(r00,vftabscale);
 222             vfitab           = _mm_cvttps_epi32(rt);
 223             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 224             vfitab           = _mm_slli_epi32(vfitab,3);
 225
 226             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 227             isaprod          = _mm_mul_ps(isai0,isaj0);
 228             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 229             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 230
 231             /* Calculate generalized born table index - this is a separate table from the normal one,
 232              * but we use the same procedure by multiplying r with scale and truncating to integer.
 233              */
 234             rt               = _mm_mul_ps(r00,gbscale);
 235             gbitab           = _mm_cvttps_epi32(rt);
 236             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 237             gbitab           = _mm_slli_epi32(gbitab,2);
 238
 239             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 240             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 241             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 242             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 243             _MM_TRANSPOSE4_PS(Y,F,G,H);
 244             Heps             = _mm_mul_ps(gbeps,H);
 245             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 246             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 247             vgb              = _mm_mul_ps(gbqqfactor,VV);
 248
 249             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 250             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 251             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 252             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 253             fjptrA           = dvda+jnrA;
 254             fjptrB           = dvda+jnrB;
 255             fjptrC           = dvda+jnrC;
 256             fjptrD           = dvda+jnrD;
 257             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 258             velec            = _mm_mul_ps(qq00,rinv00);
 259             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 260
 261             /* CUBIC SPLINE TABLE DISPERSION */
 262             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 263             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 264             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 265             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 266             _MM_TRANSPOSE4_PS(Y,F,G,H);
 267             Heps             = _mm_mul_ps(vfeps,H);
 268             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 269             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 270             vvdw6            = _mm_mul_ps(c6_00,VV);
 271             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 272             fvdw6            = _mm_mul_ps(c6_00,FF);
 273
 274             /* CUBIC SPLINE TABLE REPULSION */
 275             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm_mul_ps(c12_00,VV);
 285             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 286             fvdw12           = _mm_mul_ps(c12_00,FF);
 287             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 288             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             velecsum         = _mm_add_ps(velecsum,velec);
 292             vgbsum           = _mm_add_ps(vgbsum,vgb);
 293             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 294
 295             fscal            = _mm_add_ps(felec,fvdw);
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm_mul_ps(fscal,dx00);
 299             ty               = _mm_mul_ps(fscal,dy00);
 300             tz               = _mm_mul_ps(fscal,dz00);
 301
 302             /* Update vectorial force */
 303             fix0             = _mm_add_ps(fix0,tx);
 304             fiy0             = _mm_add_ps(fiy0,ty);
 305             fiz0             = _mm_add_ps(fiz0,tz);
 306
 307             fjptrA             = f+j_coord_offsetA;
 308             fjptrB             = f+j_coord_offsetB;
 309             fjptrC             = f+j_coord_offsetC;
 310             fjptrD             = f+j_coord_offsetD;
 311             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 312
 313             /* Inner loop uses 92 flops */
 314         }
 315
 316         if(jidx<j_index_end)
 317         {
 318
 319             /* Get j neighbor index, and coordinate index */
 320             jnrlistA         = jjnr[jidx];
 321             jnrlistB         = jjnr[jidx+1];
 322             jnrlistC         = jjnr[jidx+2];
 323             jnrlistD         = jjnr[jidx+3];
 324             /* Sign of each element will be negative for non-real atoms.
 325              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 326              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 327              */
 328             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 329             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 330             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 331             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 332             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 333             j_coord_offsetA  = DIM*jnrA;
 334             j_coord_offsetB  = DIM*jnrB;
 335             j_coord_offsetC  = DIM*jnrC;
 336             j_coord_offsetD  = DIM*jnrD;
 337
 338             /* load j atom coordinates */
 339             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 340                                               x+j_coord_offsetC,x+j_coord_offsetD,
 341                                               &jx0,&jy0,&jz0);
 342
 343             /* Calculate displacement vector */
 344             dx00             = _mm_sub_ps(ix0,jx0);
 345             dy00             = _mm_sub_ps(iy0,jy0);
 346             dz00             = _mm_sub_ps(iz0,jz0);
 347
 348             /* Calculate squared distance and things based on it */
 349             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 350
 351             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 352
 353             /* Load parameters for j particles */
 354             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 355                                                               charge+jnrC+0,charge+jnrD+0);
 356             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 357                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 358             vdwjidx0A        = 2*vdwtype[jnrA+0];
 359             vdwjidx0B        = 2*vdwtype[jnrB+0];
 360             vdwjidx0C        = 2*vdwtype[jnrC+0];
 361             vdwjidx0D        = 2*vdwtype[jnrD+0];
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             r00              = _mm_mul_ps(rsq00,rinv00);
 368             r00              = _mm_andnot_ps(dummy_mask,r00);
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq00             = _mm_mul_ps(iq0,jq0);
 372             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 373                                          vdwparam+vdwioffset0+vdwjidx0B,
 374                                          vdwparam+vdwioffset0+vdwjidx0C,
 375                                          vdwparam+vdwioffset0+vdwjidx0D,
 376                                          &c6_00,&c12_00);
 377
 378             /* Calculate table index by multiplying r with table scale and truncate to integer */
 379             rt               = _mm_mul_ps(r00,vftabscale);
 380             vfitab           = _mm_cvttps_epi32(rt);
 381             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 382             vfitab           = _mm_slli_epi32(vfitab,3);
 383
 384             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 385             isaprod          = _mm_mul_ps(isai0,isaj0);
 386             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 387             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 388
 389             /* Calculate generalized born table index - this is a separate table from the normal one,
 390              * but we use the same procedure by multiplying r with scale and truncating to integer.
 391              */
 392             rt               = _mm_mul_ps(r00,gbscale);
 393             gbitab           = _mm_cvttps_epi32(rt);
 394             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 395             gbitab           = _mm_slli_epi32(gbitab,2);
 396
 397             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 398             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 399             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 400             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 401             _MM_TRANSPOSE4_PS(Y,F,G,H);
 402             Heps             = _mm_mul_ps(gbeps,H);
 403             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 404             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 405             vgb              = _mm_mul_ps(gbqqfactor,VV);
 406
 407             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 408             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 409             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 410             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 411             /* 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. */
 412             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 413             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 414             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 415             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 416             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 417             velec            = _mm_mul_ps(qq00,rinv00);
 418             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 419
 420             /* CUBIC SPLINE TABLE DISPERSION */
 421             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 422             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 423             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 424             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 425             _MM_TRANSPOSE4_PS(Y,F,G,H);
 426             Heps             = _mm_mul_ps(vfeps,H);
 427             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 428             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 429             vvdw6            = _mm_mul_ps(c6_00,VV);
 430             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 431             fvdw6            = _mm_mul_ps(c6_00,FF);
 432
 433             /* CUBIC SPLINE TABLE REPULSION */
 434             vfitab           = _mm_add_epi32(vfitab,ifour);
 435             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 436             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 437             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 438             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 439             _MM_TRANSPOSE4_PS(Y,F,G,H);
 440             Heps             = _mm_mul_ps(vfeps,H);
 441             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 442             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 443             vvdw12           = _mm_mul_ps(c12_00,VV);
 444             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 445             fvdw12           = _mm_mul_ps(c12_00,FF);
 446             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 447             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 448
 449             /* Update potential sum for this i atom from the interaction with this j atom. */
 450             velec            = _mm_andnot_ps(dummy_mask,velec);
 451             velecsum         = _mm_add_ps(velecsum,velec);
 452             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 453             vgbsum           = _mm_add_ps(vgbsum,vgb);
 454             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 455             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 456
 457             fscal            = _mm_add_ps(felec,fvdw);
 458
 459             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 460
 461             /* Calculate temporary vectorial force */
 462             tx               = _mm_mul_ps(fscal,dx00);
 463             ty               = _mm_mul_ps(fscal,dy00);
 464             tz               = _mm_mul_ps(fscal,dz00);
 465
 466             /* Update vectorial force */
 467             fix0             = _mm_add_ps(fix0,tx);
 468             fiy0             = _mm_add_ps(fiy0,ty);
 469             fiz0             = _mm_add_ps(fiz0,tz);
 470
 471             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 472             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 473             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 474             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 475             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 476
 477             /* Inner loop uses 93 flops */
 478         }
 479
 480         /* End of innermost loop */
 481
 482         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 483                                               f+i_coord_offset,fshift+i_shift_offset);
 484
 485         ggid                        = gid[iidx];
 486         /* Update potential energies */
 487         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 488         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 489         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 490         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 491         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 492
 493         /* Increment number of inner iterations */
 494         inneriter                  += j_index_end - j_index_start;
 495
 496         /* Outer loop uses 10 flops */
 497     }
 498
 499     /* Increment number of outer iterations */
 500     outeriter        += nri;
 501
 502     /* Update outer/inner flops */
 503
 504     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*93);
 505 }
 506 /*
 507  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_single
 508  * Electrostatics interaction: GeneralizedBorn
 509  * VdW interaction:            CubicSplineTable
 510  * Geometry:                   Particle-Particle
 511  * Calculate force/pot:        Force
 512  */
 513 void
 514 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_single
 515                     (t_nblist * gmx_restrict                nlist,
 516                      rvec * gmx_restrict                    xx,
 517                      rvec * gmx_restrict                    ff,
 518                      t_forcerec * gmx_restrict              fr,
 519                      t_mdatoms * gmx_restrict               mdatoms,
 520                      nb_kernel_data_t * gmx_restrict        kernel_data,
 521                      t_nrnb * gmx_restrict                  nrnb)
 522 {
 523     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 524      * just 0 for non-waters.
 525      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 526      * jnr indices corresponding to data put in the four positions in the SIMD register.
 527      */
 528     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 529     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 530     int              jnrA,jnrB,jnrC,jnrD;
 531     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 532     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 533     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 534     real             rcutoff_scalar;
 535     real             *shiftvec,*fshift,*x,*f;
 536     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 537     real             scratch[4*DIM];
 538     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 539     int              vdwioffset0;
 540     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 541     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 542     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 543     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 544     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 545     real             *charge;
 546     __m128i          gbitab;
 547     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 548     __m128           minushalf = _mm_set1_ps(-0.5);
 549     real             *invsqrta,*dvda,*gbtab;
 550     int              nvdwtype;
 551     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 552     int              *vdwtype;
 553     real             *vdwparam;
 554     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 555     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 556     __m128i          vfitab;
 557     __m128i          ifour       = _mm_set1_epi32(4);
 558     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 559     real             *vftab;
 560     __m128           dummy_mask,cutoff_mask;
 561     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 562     __m128           one     = _mm_set1_ps(1.0);
 563     __m128           two     = _mm_set1_ps(2.0);
 564     x                = xx[0];
 565     f                = ff[0];
 566
 567     nri              = nlist->nri;
 568     iinr             = nlist->iinr;
 569     jindex           = nlist->jindex;
 570     jjnr             = nlist->jjnr;
 571     shiftidx         = nlist->shift;
 572     gid              = nlist->gid;
 573     shiftvec         = fr->shift_vec[0];
 574     fshift           = fr->fshift[0];
 575     facel            = _mm_set1_ps(fr->epsfac);
 576     charge           = mdatoms->chargeA;
 577     nvdwtype         = fr->ntype;
 578     vdwparam         = fr->nbfp;
 579     vdwtype          = mdatoms->typeA;
 580
 581     vftab            = kernel_data->table_vdw->data;
 582     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 583
 584     invsqrta         = fr->invsqrta;
 585     dvda             = fr->dvda;
 586     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 587     gbtab            = fr->gbtab.data;
 588     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 589
 590     /* Avoid stupid compiler warnings */
 591     jnrA = jnrB = jnrC = jnrD = 0;
 592     j_coord_offsetA = 0;
 593     j_coord_offsetB = 0;
 594     j_coord_offsetC = 0;
 595     j_coord_offsetD = 0;
 596
 597     outeriter        = 0;
 598     inneriter        = 0;
 599
 600     for(iidx=0;iidx<4*DIM;iidx++)
 601     {
 602         scratch[iidx] = 0.0;
 603     }
 604
 605     /* Start outer loop over neighborlists */
 606     for(iidx=0; iidx<nri; iidx++)
 607     {
 608         /* Load shift vector for this list */
 609         i_shift_offset   = DIM*shiftidx[iidx];
 610
 611         /* Load limits for loop over neighbors */
 612         j_index_start    = jindex[iidx];
 613         j_index_end      = jindex[iidx+1];
 614
 615         /* Get outer coordinate index */
 616         inr              = iinr[iidx];
 617         i_coord_offset   = DIM*inr;
 618
 619         /* Load i particle coords and add shift vector */
 620         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 621
 622         fix0             = _mm_setzero_ps();
 623         fiy0             = _mm_setzero_ps();
 624         fiz0             = _mm_setzero_ps();
 625
 626         /* Load parameters for i particles */
 627         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 628         isai0            = _mm_load1_ps(invsqrta+inr+0);
 629         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 630
 631         dvdasum          = _mm_setzero_ps();
 632
 633         /* Start inner kernel loop */
 634         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 635         {
 636
 637             /* Get j neighbor index, and coordinate index */
 638             jnrA             = jjnr[jidx];
 639             jnrB             = jjnr[jidx+1];
 640             jnrC             = jjnr[jidx+2];
 641             jnrD             = jjnr[jidx+3];
 642             j_coord_offsetA  = DIM*jnrA;
 643             j_coord_offsetB  = DIM*jnrB;
 644             j_coord_offsetC  = DIM*jnrC;
 645             j_coord_offsetD  = DIM*jnrD;
 646
 647             /* load j atom coordinates */
 648             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 649                                               x+j_coord_offsetC,x+j_coord_offsetD,
 650                                               &jx0,&jy0,&jz0);
 651
 652             /* Calculate displacement vector */
 653             dx00             = _mm_sub_ps(ix0,jx0);
 654             dy00             = _mm_sub_ps(iy0,jy0);
 655             dz00             = _mm_sub_ps(iz0,jz0);
 656
 657             /* Calculate squared distance and things based on it */
 658             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 659
 660             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 661
 662             /* Load parameters for j particles */
 663             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 664                                                               charge+jnrC+0,charge+jnrD+0);
 665             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 666                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 667             vdwjidx0A        = 2*vdwtype[jnrA+0];
 668             vdwjidx0B        = 2*vdwtype[jnrB+0];
 669             vdwjidx0C        = 2*vdwtype[jnrC+0];
 670             vdwjidx0D        = 2*vdwtype[jnrD+0];
 671
 672             /**************************
 673              * CALCULATE INTERACTIONS *
 674              **************************/
 675
 676             r00              = _mm_mul_ps(rsq00,rinv00);
 677
 678             /* Compute parameters for interactions between i and j atoms */
 679             qq00             = _mm_mul_ps(iq0,jq0);
 680             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 681                                          vdwparam+vdwioffset0+vdwjidx0B,
 682                                          vdwparam+vdwioffset0+vdwjidx0C,
 683                                          vdwparam+vdwioffset0+vdwjidx0D,
 684                                          &c6_00,&c12_00);
 685
 686             /* Calculate table index by multiplying r with table scale and truncate to integer */
 687             rt               = _mm_mul_ps(r00,vftabscale);
 688             vfitab           = _mm_cvttps_epi32(rt);
 689             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 690             vfitab           = _mm_slli_epi32(vfitab,3);
 691
 692             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 693             isaprod          = _mm_mul_ps(isai0,isaj0);
 694             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 695             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 696
 697             /* Calculate generalized born table index - this is a separate table from the normal one,
 698              * but we use the same procedure by multiplying r with scale and truncating to integer.
 699              */
 700             rt               = _mm_mul_ps(r00,gbscale);
 701             gbitab           = _mm_cvttps_epi32(rt);
 702             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 703             gbitab           = _mm_slli_epi32(gbitab,2);
 704
 705             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 706             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 707             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 708             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 709             _MM_TRANSPOSE4_PS(Y,F,G,H);
 710             Heps             = _mm_mul_ps(gbeps,H);
 711             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 712             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 713             vgb              = _mm_mul_ps(gbqqfactor,VV);
 714
 715             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 716             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 717             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 718             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 719             fjptrA           = dvda+jnrA;
 720             fjptrB           = dvda+jnrB;
 721             fjptrC           = dvda+jnrC;
 722             fjptrD           = dvda+jnrD;
 723             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 724             velec            = _mm_mul_ps(qq00,rinv00);
 725             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 726
 727             /* CUBIC SPLINE TABLE DISPERSION */
 728             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 729             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 730             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 731             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 732             _MM_TRANSPOSE4_PS(Y,F,G,H);
 733             Heps             = _mm_mul_ps(vfeps,H);
 734             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 735             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 736             fvdw6            = _mm_mul_ps(c6_00,FF);
 737
 738             /* CUBIC SPLINE TABLE REPULSION */
 739             vfitab           = _mm_add_epi32(vfitab,ifour);
 740             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 741             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 742             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 743             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 744             _MM_TRANSPOSE4_PS(Y,F,G,H);
 745             Heps             = _mm_mul_ps(vfeps,H);
 746             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 747             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 748             fvdw12           = _mm_mul_ps(c12_00,FF);
 749             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 750
 751             fscal            = _mm_add_ps(felec,fvdw);
 752
 753             /* Calculate temporary vectorial force */
 754             tx               = _mm_mul_ps(fscal,dx00);
 755             ty               = _mm_mul_ps(fscal,dy00);
 756             tz               = _mm_mul_ps(fscal,dz00);
 757
 758             /* Update vectorial force */
 759             fix0             = _mm_add_ps(fix0,tx);
 760             fiy0             = _mm_add_ps(fiy0,ty);
 761             fiz0             = _mm_add_ps(fiz0,tz);
 762
 763             fjptrA             = f+j_coord_offsetA;
 764             fjptrB             = f+j_coord_offsetB;
 765             fjptrC             = f+j_coord_offsetC;
 766             fjptrD             = f+j_coord_offsetD;
 767             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 768
 769             /* Inner loop uses 82 flops */
 770         }
 771
 772         if(jidx<j_index_end)
 773         {
 774
 775             /* Get j neighbor index, and coordinate index */
 776             jnrlistA         = jjnr[jidx];
 777             jnrlistB         = jjnr[jidx+1];
 778             jnrlistC         = jjnr[jidx+2];
 779             jnrlistD         = jjnr[jidx+3];
 780             /* Sign of each element will be negative for non-real atoms.
 781              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 782              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 783              */
 784             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 785             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 786             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 787             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 788             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 789             j_coord_offsetA  = DIM*jnrA;
 790             j_coord_offsetB  = DIM*jnrB;
 791             j_coord_offsetC  = DIM*jnrC;
 792             j_coord_offsetD  = DIM*jnrD;
 793
 794             /* load j atom coordinates */
 795             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 796                                               x+j_coord_offsetC,x+j_coord_offsetD,
 797                                               &jx0,&jy0,&jz0);
 798
 799             /* Calculate displacement vector */
 800             dx00             = _mm_sub_ps(ix0,jx0);
 801             dy00             = _mm_sub_ps(iy0,jy0);
 802             dz00             = _mm_sub_ps(iz0,jz0);
 803
 804             /* Calculate squared distance and things based on it */
 805             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 806
 807             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 808
 809             /* Load parameters for j particles */
 810             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 811                                                               charge+jnrC+0,charge+jnrD+0);
 812             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 813                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 814             vdwjidx0A        = 2*vdwtype[jnrA+0];
 815             vdwjidx0B        = 2*vdwtype[jnrB+0];
 816             vdwjidx0C        = 2*vdwtype[jnrC+0];
 817             vdwjidx0D        = 2*vdwtype[jnrD+0];
 818
 819             /**************************
 820              * CALCULATE INTERACTIONS *
 821              **************************/
 822
 823             r00              = _mm_mul_ps(rsq00,rinv00);
 824             r00              = _mm_andnot_ps(dummy_mask,r00);
 825
 826             /* Compute parameters for interactions between i and j atoms */
 827             qq00             = _mm_mul_ps(iq0,jq0);
 828             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 829                                          vdwparam+vdwioffset0+vdwjidx0B,
 830                                          vdwparam+vdwioffset0+vdwjidx0C,
 831                                          vdwparam+vdwioffset0+vdwjidx0D,
 832                                          &c6_00,&c12_00);
 833
 834             /* Calculate table index by multiplying r with table scale and truncate to integer */
 835             rt               = _mm_mul_ps(r00,vftabscale);
 836             vfitab           = _mm_cvttps_epi32(rt);
 837             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 838             vfitab           = _mm_slli_epi32(vfitab,3);
 839
 840             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 841             isaprod          = _mm_mul_ps(isai0,isaj0);
 842             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 843             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 844
 845             /* Calculate generalized born table index - this is a separate table from the normal one,
 846              * but we use the same procedure by multiplying r with scale and truncating to integer.
 847              */
 848             rt               = _mm_mul_ps(r00,gbscale);
 849             gbitab           = _mm_cvttps_epi32(rt);
 850             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 851             gbitab           = _mm_slli_epi32(gbitab,2);
 852
 853             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 854             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 855             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 856             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 857             _MM_TRANSPOSE4_PS(Y,F,G,H);
 858             Heps             = _mm_mul_ps(gbeps,H);
 859             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 860             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 861             vgb              = _mm_mul_ps(gbqqfactor,VV);
 862
 863             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 864             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 865             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 866             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 867             /* 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. */
 868             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 869             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 870             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 871             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 872             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 873             velec            = _mm_mul_ps(qq00,rinv00);
 874             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 875
 876             /* CUBIC SPLINE TABLE DISPERSION */
 877             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 878             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 879             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 880             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 881             _MM_TRANSPOSE4_PS(Y,F,G,H);
 882             Heps             = _mm_mul_ps(vfeps,H);
 883             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 884             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 885             fvdw6            = _mm_mul_ps(c6_00,FF);
 886
 887             /* CUBIC SPLINE TABLE REPULSION */
 888             vfitab           = _mm_add_epi32(vfitab,ifour);
 889             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 890             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 891             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 892             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 893             _MM_TRANSPOSE4_PS(Y,F,G,H);
 894             Heps             = _mm_mul_ps(vfeps,H);
 895             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 896             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 897             fvdw12           = _mm_mul_ps(c12_00,FF);
 898             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 899
 900             fscal            = _mm_add_ps(felec,fvdw);
 901
 902             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 903
 904             /* Calculate temporary vectorial force */
 905             tx               = _mm_mul_ps(fscal,dx00);
 906             ty               = _mm_mul_ps(fscal,dy00);
 907             tz               = _mm_mul_ps(fscal,dz00);
 908
 909             /* Update vectorial force */
 910             fix0             = _mm_add_ps(fix0,tx);
 911             fiy0             = _mm_add_ps(fiy0,ty);
 912             fiz0             = _mm_add_ps(fiz0,tz);
 913
 914             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 915             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 916             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 917             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 918             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 919
 920             /* Inner loop uses 83 flops */
 921         }
 922
 923         /* End of innermost loop */
 924
 925         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 926                                               f+i_coord_offset,fshift+i_shift_offset);
 927
 928         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 929         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 930
 931         /* Increment number of inner iterations */
 932         inneriter                  += j_index_end - j_index_start;
 933
 934         /* Outer loop uses 7 flops */
 935     }
 936
 937     /* Increment number of outer iterations */
 938     outeriter        += nri;
 939
 940     /* Update outer/inner flops */
 941
 942     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);
 943 }