src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecGB_VdwCSTab_GeomP1P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse4_1_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_sse4_1_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_round_ps(rt, _MM_FROUND_FLOOR));
 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_round_ps(rt, _MM_FROUND_FLOOR));
 237             gbitab           = _mm_slli_epi32(gbitab,2);
 238             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 239             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 240             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 241             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 242             _MM_TRANSPOSE4_PS(Y,F,G,H);
 243             Heps             = _mm_mul_ps(gbeps,H);
 244             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 245             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 246             vgb              = _mm_mul_ps(gbqqfactor,VV);
 247
 248             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 249             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 250             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 251             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 252             fjptrA           = dvda+jnrA;
 253             fjptrB           = dvda+jnrB;
 254             fjptrC           = dvda+jnrC;
 255             fjptrD           = dvda+jnrD;
 256             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 257             velec            = _mm_mul_ps(qq00,rinv00);
 258             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 259
 260             /* CUBIC SPLINE TABLE DISPERSION */
 261             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 262             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 263             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 264             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 265             _MM_TRANSPOSE4_PS(Y,F,G,H);
 266             Heps             = _mm_mul_ps(vfeps,H);
 267             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 268             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 269             vvdw6            = _mm_mul_ps(c6_00,VV);
 270             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 271             fvdw6            = _mm_mul_ps(c6_00,FF);
 272
 273             /* CUBIC SPLINE TABLE REPULSION */
 274             vfitab           = _mm_add_epi32(vfitab,ifour);
 275             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 276             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 277             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 278             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 279             _MM_TRANSPOSE4_PS(Y,F,G,H);
 280             Heps             = _mm_mul_ps(vfeps,H);
 281             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 282             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 283             vvdw12           = _mm_mul_ps(c12_00,VV);
 284             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 285             fvdw12           = _mm_mul_ps(c12_00,FF);
 286             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 287             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             velecsum         = _mm_add_ps(velecsum,velec);
 291             vgbsum           = _mm_add_ps(vgbsum,vgb);
 292             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 293
 294             fscal            = _mm_add_ps(felec,fvdw);
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm_mul_ps(fscal,dx00);
 298             ty               = _mm_mul_ps(fscal,dy00);
 299             tz               = _mm_mul_ps(fscal,dz00);
 300
 301             /* Update vectorial force */
 302             fix0             = _mm_add_ps(fix0,tx);
 303             fiy0             = _mm_add_ps(fiy0,ty);
 304             fiz0             = _mm_add_ps(fiz0,tz);
 305
 306             fjptrA             = f+j_coord_offsetA;
 307             fjptrB             = f+j_coord_offsetB;
 308             fjptrC             = f+j_coord_offsetC;
 309             fjptrD             = f+j_coord_offsetD;
 310             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 311
 312             /* Inner loop uses 92 flops */
 313         }
 314
 315         if(jidx<j_index_end)
 316         {
 317
 318             /* Get j neighbor index, and coordinate index */
 319             jnrlistA         = jjnr[jidx];
 320             jnrlistB         = jjnr[jidx+1];
 321             jnrlistC         = jjnr[jidx+2];
 322             jnrlistD         = jjnr[jidx+3];
 323             /* Sign of each element will be negative for non-real atoms.
 324              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 325              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 326              */
 327             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 328             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 329             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 330             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 331             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 332             j_coord_offsetA  = DIM*jnrA;
 333             j_coord_offsetB  = DIM*jnrB;
 334             j_coord_offsetC  = DIM*jnrC;
 335             j_coord_offsetD  = DIM*jnrD;
 336
 337             /* load j atom coordinates */
 338             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 339                                               x+j_coord_offsetC,x+j_coord_offsetD,
 340                                               &jx0,&jy0,&jz0);
 341
 342             /* Calculate displacement vector */
 343             dx00             = _mm_sub_ps(ix0,jx0);
 344             dy00             = _mm_sub_ps(iy0,jy0);
 345             dz00             = _mm_sub_ps(iz0,jz0);
 346
 347             /* Calculate squared distance and things based on it */
 348             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 349
 350             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 351
 352             /* Load parameters for j particles */
 353             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 354                                                               charge+jnrC+0,charge+jnrD+0);
 355             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 356                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 357             vdwjidx0A        = 2*vdwtype[jnrA+0];
 358             vdwjidx0B        = 2*vdwtype[jnrB+0];
 359             vdwjidx0C        = 2*vdwtype[jnrC+0];
 360             vdwjidx0D        = 2*vdwtype[jnrD+0];
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             r00              = _mm_mul_ps(rsq00,rinv00);
 367             r00              = _mm_andnot_ps(dummy_mask,r00);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq00             = _mm_mul_ps(iq0,jq0);
 371             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 372                                          vdwparam+vdwioffset0+vdwjidx0B,
 373                                          vdwparam+vdwioffset0+vdwjidx0C,
 374                                          vdwparam+vdwioffset0+vdwjidx0D,
 375                                          &c6_00,&c12_00);
 376
 377             /* Calculate table index by multiplying r with table scale and truncate to integer */
 378             rt               = _mm_mul_ps(r00,vftabscale);
 379             vfitab           = _mm_cvttps_epi32(rt);
 380             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 381             vfitab           = _mm_slli_epi32(vfitab,3);
 382
 383             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 384             isaprod          = _mm_mul_ps(isai0,isaj0);
 385             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 386             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 387
 388             /* Calculate generalized born table index - this is a separate table from the normal one,
 389              * but we use the same procedure by multiplying r with scale and truncating to integer.
 390              */
 391             rt               = _mm_mul_ps(r00,gbscale);
 392             gbitab           = _mm_cvttps_epi32(rt);
 393             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 394             gbitab           = _mm_slli_epi32(gbitab,2);
 395             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 396             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 397             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 398             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 399             _MM_TRANSPOSE4_PS(Y,F,G,H);
 400             Heps             = _mm_mul_ps(gbeps,H);
 401             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 402             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 403             vgb              = _mm_mul_ps(gbqqfactor,VV);
 404
 405             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 406             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 407             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 408             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 409             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 410             /* 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. */
 411             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 412             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 413             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 414             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 415             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 416             velec            = _mm_mul_ps(qq00,rinv00);
 417             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 418
 419             /* CUBIC SPLINE TABLE DISPERSION */
 420             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 421             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 422             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 423             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 424             _MM_TRANSPOSE4_PS(Y,F,G,H);
 425             Heps             = _mm_mul_ps(vfeps,H);
 426             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 427             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 428             vvdw6            = _mm_mul_ps(c6_00,VV);
 429             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 430             fvdw6            = _mm_mul_ps(c6_00,FF);
 431
 432             /* CUBIC SPLINE TABLE REPULSION */
 433             vfitab           = _mm_add_epi32(vfitab,ifour);
 434             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 435             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 436             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 437             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 438             _MM_TRANSPOSE4_PS(Y,F,G,H);
 439             Heps             = _mm_mul_ps(vfeps,H);
 440             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 441             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 442             vvdw12           = _mm_mul_ps(c12_00,VV);
 443             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 444             fvdw12           = _mm_mul_ps(c12_00,FF);
 445             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 446             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 447
 448             /* Update potential sum for this i atom from the interaction with this j atom. */
 449             velec            = _mm_andnot_ps(dummy_mask,velec);
 450             velecsum         = _mm_add_ps(velecsum,velec);
 451             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 452             vgbsum           = _mm_add_ps(vgbsum,vgb);
 453             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 454             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 455
 456             fscal            = _mm_add_ps(felec,fvdw);
 457
 458             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 459
 460             /* Calculate temporary vectorial force */
 461             tx               = _mm_mul_ps(fscal,dx00);
 462             ty               = _mm_mul_ps(fscal,dy00);
 463             tz               = _mm_mul_ps(fscal,dz00);
 464
 465             /* Update vectorial force */
 466             fix0             = _mm_add_ps(fix0,tx);
 467             fiy0             = _mm_add_ps(fiy0,ty);
 468             fiz0             = _mm_add_ps(fiz0,tz);
 469
 470             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 471             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 472             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 473             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 474             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 475
 476             /* Inner loop uses 93 flops */
 477         }
 478
 479         /* End of innermost loop */
 480
 481         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 482                                               f+i_coord_offset,fshift+i_shift_offset);
 483
 484         ggid                        = gid[iidx];
 485         /* Update potential energies */
 486         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 487         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 488         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 489         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 490         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 491
 492         /* Increment number of inner iterations */
 493         inneriter                  += j_index_end - j_index_start;
 494
 495         /* Outer loop uses 10 flops */
 496     }
 497
 498     /* Increment number of outer iterations */
 499     outeriter        += nri;
 500
 501     /* Update outer/inner flops */
 502
 503     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*93);
 504 }
 505 /*
 506  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse4_1_single
 507  * Electrostatics interaction: GeneralizedBorn
 508  * VdW interaction:            CubicSplineTable
 509  * Geometry:                   Particle-Particle
 510  * Calculate force/pot:        Force
 511  */
 512 void
 513 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse4_1_single
 514                     (t_nblist * gmx_restrict                nlist,
 515                      rvec * gmx_restrict                    xx,
 516                      rvec * gmx_restrict                    ff,
 517                      t_forcerec * gmx_restrict              fr,
 518                      t_mdatoms * gmx_restrict               mdatoms,
 519                      nb_kernel_data_t * gmx_restrict        kernel_data,
 520                      t_nrnb * gmx_restrict                  nrnb)
 521 {
 522     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 523      * just 0 for non-waters.
 524      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 525      * jnr indices corresponding to data put in the four positions in the SIMD register.
 526      */
 527     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 528     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 529     int              jnrA,jnrB,jnrC,jnrD;
 530     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 531     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 532     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 533     real             rcutoff_scalar;
 534     real             *shiftvec,*fshift,*x,*f;
 535     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 536     real             scratch[4*DIM];
 537     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 538     int              vdwioffset0;
 539     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 540     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 541     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 542     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 543     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 544     real             *charge;
 545     __m128i          gbitab;
 546     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 547     __m128           minushalf = _mm_set1_ps(-0.5);
 548     real             *invsqrta,*dvda,*gbtab;
 549     int              nvdwtype;
 550     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 551     int              *vdwtype;
 552     real             *vdwparam;
 553     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 554     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 555     __m128i          vfitab;
 556     __m128i          ifour       = _mm_set1_epi32(4);
 557     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 558     real             *vftab;
 559     __m128           dummy_mask,cutoff_mask;
 560     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 561     __m128           one     = _mm_set1_ps(1.0);
 562     __m128           two     = _mm_set1_ps(2.0);
 563     x                = xx[0];
 564     f                = ff[0];
 565
 566     nri              = nlist->nri;
 567     iinr             = nlist->iinr;
 568     jindex           = nlist->jindex;
 569     jjnr             = nlist->jjnr;
 570     shiftidx         = nlist->shift;
 571     gid              = nlist->gid;
 572     shiftvec         = fr->shift_vec[0];
 573     fshift           = fr->fshift[0];
 574     facel            = _mm_set1_ps(fr->epsfac);
 575     charge           = mdatoms->chargeA;
 576     nvdwtype         = fr->ntype;
 577     vdwparam         = fr->nbfp;
 578     vdwtype          = mdatoms->typeA;
 579
 580     vftab            = kernel_data->table_vdw->data;
 581     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 582
 583     invsqrta         = fr->invsqrta;
 584     dvda             = fr->dvda;
 585     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 586     gbtab            = fr->gbtab.data;
 587     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 588
 589     /* Avoid stupid compiler warnings */
 590     jnrA = jnrB = jnrC = jnrD = 0;
 591     j_coord_offsetA = 0;
 592     j_coord_offsetB = 0;
 593     j_coord_offsetC = 0;
 594     j_coord_offsetD = 0;
 595
 596     outeriter        = 0;
 597     inneriter        = 0;
 598
 599     for(iidx=0;iidx<4*DIM;iidx++)
 600     {
 601         scratch[iidx] = 0.0;
 602     }
 603
 604     /* Start outer loop over neighborlists */
 605     for(iidx=0; iidx<nri; iidx++)
 606     {
 607         /* Load shift vector for this list */
 608         i_shift_offset   = DIM*shiftidx[iidx];
 609
 610         /* Load limits for loop over neighbors */
 611         j_index_start    = jindex[iidx];
 612         j_index_end      = jindex[iidx+1];
 613
 614         /* Get outer coordinate index */
 615         inr              = iinr[iidx];
 616         i_coord_offset   = DIM*inr;
 617
 618         /* Load i particle coords and add shift vector */
 619         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 620
 621         fix0             = _mm_setzero_ps();
 622         fiy0             = _mm_setzero_ps();
 623         fiz0             = _mm_setzero_ps();
 624
 625         /* Load parameters for i particles */
 626         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 627         isai0            = _mm_load1_ps(invsqrta+inr+0);
 628         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 629
 630         dvdasum          = _mm_setzero_ps();
 631
 632         /* Start inner kernel loop */
 633         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 634         {
 635
 636             /* Get j neighbor index, and coordinate index */
 637             jnrA             = jjnr[jidx];
 638             jnrB             = jjnr[jidx+1];
 639             jnrC             = jjnr[jidx+2];
 640             jnrD             = jjnr[jidx+3];
 641             j_coord_offsetA  = DIM*jnrA;
 642             j_coord_offsetB  = DIM*jnrB;
 643             j_coord_offsetC  = DIM*jnrC;
 644             j_coord_offsetD  = DIM*jnrD;
 645
 646             /* load j atom coordinates */
 647             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 648                                               x+j_coord_offsetC,x+j_coord_offsetD,
 649                                               &jx0,&jy0,&jz0);
 650
 651             /* Calculate displacement vector */
 652             dx00             = _mm_sub_ps(ix0,jx0);
 653             dy00             = _mm_sub_ps(iy0,jy0);
 654             dz00             = _mm_sub_ps(iz0,jz0);
 655
 656             /* Calculate squared distance and things based on it */
 657             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 658
 659             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 660
 661             /* Load parameters for j particles */
 662             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 663                                                               charge+jnrC+0,charge+jnrD+0);
 664             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 665                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 666             vdwjidx0A        = 2*vdwtype[jnrA+0];
 667             vdwjidx0B        = 2*vdwtype[jnrB+0];
 668             vdwjidx0C        = 2*vdwtype[jnrC+0];
 669             vdwjidx0D        = 2*vdwtype[jnrD+0];
 670
 671             /**************************
 672              * CALCULATE INTERACTIONS *
 673              **************************/
 674
 675             r00              = _mm_mul_ps(rsq00,rinv00);
 676
 677             /* Compute parameters for interactions between i and j atoms */
 678             qq00             = _mm_mul_ps(iq0,jq0);
 679             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 680                                          vdwparam+vdwioffset0+vdwjidx0B,
 681                                          vdwparam+vdwioffset0+vdwjidx0C,
 682                                          vdwparam+vdwioffset0+vdwjidx0D,
 683                                          &c6_00,&c12_00);
 684
 685             /* Calculate table index by multiplying r with table scale and truncate to integer */
 686             rt               = _mm_mul_ps(r00,vftabscale);
 687             vfitab           = _mm_cvttps_epi32(rt);
 688             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 689             vfitab           = _mm_slli_epi32(vfitab,3);
 690
 691             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 692             isaprod          = _mm_mul_ps(isai0,isaj0);
 693             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 694             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 695
 696             /* Calculate generalized born table index - this is a separate table from the normal one,
 697              * but we use the same procedure by multiplying r with scale and truncating to integer.
 698              */
 699             rt               = _mm_mul_ps(r00,gbscale);
 700             gbitab           = _mm_cvttps_epi32(rt);
 701             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 702             gbitab           = _mm_slli_epi32(gbitab,2);
 703             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 704             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 705             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 706             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 707             _MM_TRANSPOSE4_PS(Y,F,G,H);
 708             Heps             = _mm_mul_ps(gbeps,H);
 709             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 710             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 711             vgb              = _mm_mul_ps(gbqqfactor,VV);
 712
 713             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 714             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 715             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 716             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 717             fjptrA           = dvda+jnrA;
 718             fjptrB           = dvda+jnrB;
 719             fjptrC           = dvda+jnrC;
 720             fjptrD           = dvda+jnrD;
 721             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 722             velec            = _mm_mul_ps(qq00,rinv00);
 723             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 724
 725             /* CUBIC SPLINE TABLE DISPERSION */
 726             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 727             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 728             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 729             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 730             _MM_TRANSPOSE4_PS(Y,F,G,H);
 731             Heps             = _mm_mul_ps(vfeps,H);
 732             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 733             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 734             fvdw6            = _mm_mul_ps(c6_00,FF);
 735
 736             /* CUBIC SPLINE TABLE REPULSION */
 737             vfitab           = _mm_add_epi32(vfitab,ifour);
 738             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 739             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 740             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 741             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 742             _MM_TRANSPOSE4_PS(Y,F,G,H);
 743             Heps             = _mm_mul_ps(vfeps,H);
 744             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 745             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 746             fvdw12           = _mm_mul_ps(c12_00,FF);
 747             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 748
 749             fscal            = _mm_add_ps(felec,fvdw);
 750
 751             /* Calculate temporary vectorial force */
 752             tx               = _mm_mul_ps(fscal,dx00);
 753             ty               = _mm_mul_ps(fscal,dy00);
 754             tz               = _mm_mul_ps(fscal,dz00);
 755
 756             /* Update vectorial force */
 757             fix0             = _mm_add_ps(fix0,tx);
 758             fiy0             = _mm_add_ps(fiy0,ty);
 759             fiz0             = _mm_add_ps(fiz0,tz);
 760
 761             fjptrA             = f+j_coord_offsetA;
 762             fjptrB             = f+j_coord_offsetB;
 763             fjptrC             = f+j_coord_offsetC;
 764             fjptrD             = f+j_coord_offsetD;
 765             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 766
 767             /* Inner loop uses 82 flops */
 768         }
 769
 770         if(jidx<j_index_end)
 771         {
 772
 773             /* Get j neighbor index, and coordinate index */
 774             jnrlistA         = jjnr[jidx];
 775             jnrlistB         = jjnr[jidx+1];
 776             jnrlistC         = jjnr[jidx+2];
 777             jnrlistD         = jjnr[jidx+3];
 778             /* Sign of each element will be negative for non-real atoms.
 779              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 780              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 781              */
 782             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 783             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 784             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 785             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 786             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 787             j_coord_offsetA  = DIM*jnrA;
 788             j_coord_offsetB  = DIM*jnrB;
 789             j_coord_offsetC  = DIM*jnrC;
 790             j_coord_offsetD  = DIM*jnrD;
 791
 792             /* load j atom coordinates */
 793             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 794                                               x+j_coord_offsetC,x+j_coord_offsetD,
 795                                               &jx0,&jy0,&jz0);
 796
 797             /* Calculate displacement vector */
 798             dx00             = _mm_sub_ps(ix0,jx0);
 799             dy00             = _mm_sub_ps(iy0,jy0);
 800             dz00             = _mm_sub_ps(iz0,jz0);
 801
 802             /* Calculate squared distance and things based on it */
 803             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 804
 805             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 806
 807             /* Load parameters for j particles */
 808             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 809                                                               charge+jnrC+0,charge+jnrD+0);
 810             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 811                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 812             vdwjidx0A        = 2*vdwtype[jnrA+0];
 813             vdwjidx0B        = 2*vdwtype[jnrB+0];
 814             vdwjidx0C        = 2*vdwtype[jnrC+0];
 815             vdwjidx0D        = 2*vdwtype[jnrD+0];
 816
 817             /**************************
 818              * CALCULATE INTERACTIONS *
 819              **************************/
 820
 821             r00              = _mm_mul_ps(rsq00,rinv00);
 822             r00              = _mm_andnot_ps(dummy_mask,r00);
 823
 824             /* Compute parameters for interactions between i and j atoms */
 825             qq00             = _mm_mul_ps(iq0,jq0);
 826             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 827                                          vdwparam+vdwioffset0+vdwjidx0B,
 828                                          vdwparam+vdwioffset0+vdwjidx0C,
 829                                          vdwparam+vdwioffset0+vdwjidx0D,
 830                                          &c6_00,&c12_00);
 831
 832             /* Calculate table index by multiplying r with table scale and truncate to integer */
 833             rt               = _mm_mul_ps(r00,vftabscale);
 834             vfitab           = _mm_cvttps_epi32(rt);
 835             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 836             vfitab           = _mm_slli_epi32(vfitab,3);
 837
 838             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 839             isaprod          = _mm_mul_ps(isai0,isaj0);
 840             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 841             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 842
 843             /* Calculate generalized born table index - this is a separate table from the normal one,
 844              * but we use the same procedure by multiplying r with scale and truncating to integer.
 845              */
 846             rt               = _mm_mul_ps(r00,gbscale);
 847             gbitab           = _mm_cvttps_epi32(rt);
 848             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 849             gbitab           = _mm_slli_epi32(gbitab,2);
 850             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 851             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 852             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 853             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 854             _MM_TRANSPOSE4_PS(Y,F,G,H);
 855             Heps             = _mm_mul_ps(gbeps,H);
 856             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 857             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 858             vgb              = _mm_mul_ps(gbqqfactor,VV);
 859
 860             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 861             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 862             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 863             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 864             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 865             /* 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. */
 866             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 867             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 868             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 869             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 870             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 871             velec            = _mm_mul_ps(qq00,rinv00);
 872             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 873
 874             /* CUBIC SPLINE TABLE DISPERSION */
 875             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 876             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 877             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 878             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 879             _MM_TRANSPOSE4_PS(Y,F,G,H);
 880             Heps             = _mm_mul_ps(vfeps,H);
 881             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 882             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 883             fvdw6            = _mm_mul_ps(c6_00,FF);
 884
 885             /* CUBIC SPLINE TABLE REPULSION */
 886             vfitab           = _mm_add_epi32(vfitab,ifour);
 887             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 888             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 889             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 890             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 891             _MM_TRANSPOSE4_PS(Y,F,G,H);
 892             Heps             = _mm_mul_ps(vfeps,H);
 893             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 894             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 895             fvdw12           = _mm_mul_ps(c12_00,FF);
 896             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 897
 898             fscal            = _mm_add_ps(felec,fvdw);
 899
 900             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 901
 902             /* Calculate temporary vectorial force */
 903             tx               = _mm_mul_ps(fscal,dx00);
 904             ty               = _mm_mul_ps(fscal,dy00);
 905             tz               = _mm_mul_ps(fscal,dz00);
 906
 907             /* Update vectorial force */
 908             fix0             = _mm_add_ps(fix0,tx);
 909             fiy0             = _mm_add_ps(fiy0,ty);
 910             fiz0             = _mm_add_ps(fiz0,tz);
 911
 912             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 913             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 914             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 915             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 916             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 917
 918             /* Inner loop uses 83 flops */
 919         }
 920
 921         /* End of innermost loop */
 922
 923         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 924                                               f+i_coord_offset,fshift+i_shift_offset);
 925
 926         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 927         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 928
 929         /* Increment number of inner iterations */
 930         inneriter                  += j_index_end - j_index_start;
 931
 932         /* Outer loop uses 7 flops */
 933     }
 934
 935     /* Increment number of outer iterations */
 936     outeriter        += nri;
 937
 938     /* Update outer/inner flops */
 939
 940     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*83);
 941 }