src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_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_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_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              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  94     real             *vftab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_elec_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 122     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 123     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 158
 159         fix0             = _mm_setzero_ps();
 160         fiy0             = _mm_setzero_ps();
 161         fiz0             = _mm_setzero_ps();
 162         fix1             = _mm_setzero_ps();
 163         fiy1             = _mm_setzero_ps();
 164         fiz1             = _mm_setzero_ps();
 165         fix2             = _mm_setzero_ps();
 166         fiy2             = _mm_setzero_ps();
 167         fiz2             = _mm_setzero_ps();
 168         fix3             = _mm_setzero_ps();
 169         fiy3             = _mm_setzero_ps();
 170         fiz3             = _mm_setzero_ps();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_ps();
 174         vvdwsum          = _mm_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187             j_coord_offsetC  = DIM*jnrC;
 188             j_coord_offsetD  = DIM*jnrD;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               x+j_coord_offsetC,x+j_coord_offsetD,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_ps(ix0,jx0);
 197             dy00             = _mm_sub_ps(iy0,jy0);
 198             dz00             = _mm_sub_ps(iz0,jz0);
 199             dx10             = _mm_sub_ps(ix1,jx0);
 200             dy10             = _mm_sub_ps(iy1,jy0);
 201             dz10             = _mm_sub_ps(iz1,jz0);
 202             dx20             = _mm_sub_ps(ix2,jx0);
 203             dy20             = _mm_sub_ps(iy2,jy0);
 204             dz20             = _mm_sub_ps(iz2,jz0);
 205             dx30             = _mm_sub_ps(ix3,jx0);
 206             dy30             = _mm_sub_ps(iy3,jy0);
 207             dz30             = _mm_sub_ps(iz3,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 213             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 214
 215             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 216             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 217             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 218             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 222                                                               charge+jnrC+0,charge+jnrD+0);
 223             vdwjidx0A        = 2*vdwtype[jnrA+0];
 224             vdwjidx0B        = 2*vdwtype[jnrB+0];
 225             vdwjidx0C        = 2*vdwtype[jnrC+0];
 226             vdwjidx0D        = 2*vdwtype[jnrD+0];
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             r00              = _mm_mul_ps(rsq00,rinv00);
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 236                                          vdwparam+vdwioffset0+vdwjidx0B,
 237                                          vdwparam+vdwioffset0+vdwjidx0C,
 238                                          vdwparam+vdwioffset0+vdwjidx0D,
 239                                          &c6_00,&c12_00);
 240
 241             /* Calculate table index by multiplying r with table scale and truncate to integer */
 242             rt               = _mm_mul_ps(r00,vftabscale);
 243             vfitab           = _mm_cvttps_epi32(rt);
 244             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 245             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 246
 247             /* CUBIC SPLINE TABLE DISPERSION */
 248             vfitab           = _mm_add_epi32(vfitab,ifour);
 249             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 250             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 251             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 252             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 253             _MM_TRANSPOSE4_PS(Y,F,G,H);
 254             Heps             = _mm_mul_ps(vfeps,H);
 255             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 256             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 257             vvdw6            = _mm_mul_ps(c6_00,VV);
 258             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 259             fvdw6            = _mm_mul_ps(c6_00,FF);
 260
 261             /* CUBIC SPLINE TABLE REPULSION */
 262             vfitab           = _mm_add_epi32(vfitab,ifour);
 263             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 264             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 265             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 266             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 267             _MM_TRANSPOSE4_PS(Y,F,G,H);
 268             Heps             = _mm_mul_ps(vfeps,H);
 269             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 270             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 271             vvdw12           = _mm_mul_ps(c12_00,VV);
 272             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 273             fvdw12           = _mm_mul_ps(c12_00,FF);
 274             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 275             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 279
 280             fscal            = fvdw;
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm_mul_ps(fscal,dx00);
 284             ty               = _mm_mul_ps(fscal,dy00);
 285             tz               = _mm_mul_ps(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_add_ps(fix0,tx);
 289             fiy0             = _mm_add_ps(fiy0,ty);
 290             fiz0             = _mm_add_ps(fiz0,tz);
 291
 292             fjptrA             = f+j_coord_offsetA;
 293             fjptrB             = f+j_coord_offsetB;
 294             fjptrC             = f+j_coord_offsetC;
 295             fjptrD             = f+j_coord_offsetD;
 296             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             r10              = _mm_mul_ps(rsq10,rinv10);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm_mul_ps(iq1,jq0);
 306
 307             /* Calculate table index by multiplying r with table scale and truncate to integer */
 308             rt               = _mm_mul_ps(r10,vftabscale);
 309             vfitab           = _mm_cvttps_epi32(rt);
 310             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 311             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 312
 313             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 314             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 315             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 316             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 317             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 318             _MM_TRANSPOSE4_PS(Y,F,G,H);
 319             Heps             = _mm_mul_ps(vfeps,H);
 320             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 321             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 322             velec            = _mm_mul_ps(qq10,VV);
 323             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 324             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velecsum         = _mm_add_ps(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm_mul_ps(fscal,dx10);
 333             ty               = _mm_mul_ps(fscal,dy10);
 334             tz               = _mm_mul_ps(fscal,dz10);
 335
 336             /* Update vectorial force */
 337             fix1             = _mm_add_ps(fix1,tx);
 338             fiy1             = _mm_add_ps(fiy1,ty);
 339             fiz1             = _mm_add_ps(fiz1,tz);
 340
 341             fjptrA             = f+j_coord_offsetA;
 342             fjptrB             = f+j_coord_offsetB;
 343             fjptrC             = f+j_coord_offsetC;
 344             fjptrD             = f+j_coord_offsetD;
 345             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r20              = _mm_mul_ps(rsq20,rinv20);
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm_mul_ps(iq2,jq0);
 355
 356             /* Calculate table index by multiplying r with table scale and truncate to integer */
 357             rt               = _mm_mul_ps(r20,vftabscale);
 358             vfitab           = _mm_cvttps_epi32(rt);
 359             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 360             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 361
 362             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 363             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 364             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 365             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 366             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 367             _MM_TRANSPOSE4_PS(Y,F,G,H);
 368             Heps             = _mm_mul_ps(vfeps,H);
 369             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 370             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 371             velec            = _mm_mul_ps(qq20,VV);
 372             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 373             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velecsum         = _mm_add_ps(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm_mul_ps(fscal,dx20);
 382             ty               = _mm_mul_ps(fscal,dy20);
 383             tz               = _mm_mul_ps(fscal,dz20);
 384
 385             /* Update vectorial force */
 386             fix2             = _mm_add_ps(fix2,tx);
 387             fiy2             = _mm_add_ps(fiy2,ty);
 388             fiz2             = _mm_add_ps(fiz2,tz);
 389
 390             fjptrA             = f+j_coord_offsetA;
 391             fjptrB             = f+j_coord_offsetB;
 392             fjptrC             = f+j_coord_offsetC;
 393             fjptrD             = f+j_coord_offsetD;
 394             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             r30              = _mm_mul_ps(rsq30,rinv30);
 401
 402             /* Compute parameters for interactions between i and j atoms */
 403             qq30             = _mm_mul_ps(iq3,jq0);
 404
 405             /* Calculate table index by multiplying r with table scale and truncate to integer */
 406             rt               = _mm_mul_ps(r30,vftabscale);
 407             vfitab           = _mm_cvttps_epi32(rt);
 408             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 409             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 410
 411             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 412             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 413             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 414             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 415             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 416             _MM_TRANSPOSE4_PS(Y,F,G,H);
 417             Heps             = _mm_mul_ps(vfeps,H);
 418             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 419             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 420             velec            = _mm_mul_ps(qq30,VV);
 421             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 422             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 423
 424             /* Update potential sum for this i atom from the interaction with this j atom. */
 425             velecsum         = _mm_add_ps(velecsum,velec);
 426
 427             fscal            = felec;
 428
 429             /* Calculate temporary vectorial force */
 430             tx               = _mm_mul_ps(fscal,dx30);
 431             ty               = _mm_mul_ps(fscal,dy30);
 432             tz               = _mm_mul_ps(fscal,dz30);
 433
 434             /* Update vectorial force */
 435             fix3             = _mm_add_ps(fix3,tx);
 436             fiy3             = _mm_add_ps(fiy3,ty);
 437             fiz3             = _mm_add_ps(fiz3,tz);
 438
 439             fjptrA             = f+j_coord_offsetA;
 440             fjptrB             = f+j_coord_offsetB;
 441             fjptrC             = f+j_coord_offsetC;
 442             fjptrD             = f+j_coord_offsetD;
 443             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 444
 445             /* Inner loop uses 185 flops */
 446         }
 447
 448         if(jidx<j_index_end)
 449         {
 450
 451             /* Get j neighbor index, and coordinate index */
 452             jnrlistA         = jjnr[jidx];
 453             jnrlistB         = jjnr[jidx+1];
 454             jnrlistC         = jjnr[jidx+2];
 455             jnrlistD         = jjnr[jidx+3];
 456             /* Sign of each element will be negative for non-real atoms.
 457              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 458              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 459              */
 460             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 461             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 462             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 463             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 464             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 465             j_coord_offsetA  = DIM*jnrA;
 466             j_coord_offsetB  = DIM*jnrB;
 467             j_coord_offsetC  = DIM*jnrC;
 468             j_coord_offsetD  = DIM*jnrD;
 469
 470             /* load j atom coordinates */
 471             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 472                                               x+j_coord_offsetC,x+j_coord_offsetD,
 473                                               &jx0,&jy0,&jz0);
 474
 475             /* Calculate displacement vector */
 476             dx00             = _mm_sub_ps(ix0,jx0);
 477             dy00             = _mm_sub_ps(iy0,jy0);
 478             dz00             = _mm_sub_ps(iz0,jz0);
 479             dx10             = _mm_sub_ps(ix1,jx0);
 480             dy10             = _mm_sub_ps(iy1,jy0);
 481             dz10             = _mm_sub_ps(iz1,jz0);
 482             dx20             = _mm_sub_ps(ix2,jx0);
 483             dy20             = _mm_sub_ps(iy2,jy0);
 484             dz20             = _mm_sub_ps(iz2,jz0);
 485             dx30             = _mm_sub_ps(ix3,jx0);
 486             dy30             = _mm_sub_ps(iy3,jy0);
 487             dz30             = _mm_sub_ps(iz3,jz0);
 488
 489             /* Calculate squared distance and things based on it */
 490             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 491             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 492             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 493             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 494
 495             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 496             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 497             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 498             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 499
 500             /* Load parameters for j particles */
 501             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 502                                                               charge+jnrC+0,charge+jnrD+0);
 503             vdwjidx0A        = 2*vdwtype[jnrA+0];
 504             vdwjidx0B        = 2*vdwtype[jnrB+0];
 505             vdwjidx0C        = 2*vdwtype[jnrC+0];
 506             vdwjidx0D        = 2*vdwtype[jnrD+0];
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             r00              = _mm_mul_ps(rsq00,rinv00);
 513             r00              = _mm_andnot_ps(dummy_mask,r00);
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 517                                          vdwparam+vdwioffset0+vdwjidx0B,
 518                                          vdwparam+vdwioffset0+vdwjidx0C,
 519                                          vdwparam+vdwioffset0+vdwjidx0D,
 520                                          &c6_00,&c12_00);
 521
 522             /* Calculate table index by multiplying r with table scale and truncate to integer */
 523             rt               = _mm_mul_ps(r00,vftabscale);
 524             vfitab           = _mm_cvttps_epi32(rt);
 525             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 526             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 527
 528             /* CUBIC SPLINE TABLE DISPERSION */
 529             vfitab           = _mm_add_epi32(vfitab,ifour);
 530             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 531             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 532             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 533             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 534             _MM_TRANSPOSE4_PS(Y,F,G,H);
 535             Heps             = _mm_mul_ps(vfeps,H);
 536             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 537             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 538             vvdw6            = _mm_mul_ps(c6_00,VV);
 539             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 540             fvdw6            = _mm_mul_ps(c6_00,FF);
 541
 542             /* CUBIC SPLINE TABLE REPULSION */
 543             vfitab           = _mm_add_epi32(vfitab,ifour);
 544             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 545             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 546             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 547             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 548             _MM_TRANSPOSE4_PS(Y,F,G,H);
 549             Heps             = _mm_mul_ps(vfeps,H);
 550             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 551             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 552             vvdw12           = _mm_mul_ps(c12_00,VV);
 553             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 554             fvdw12           = _mm_mul_ps(c12_00,FF);
 555             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 556             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 560             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 561
 562             fscal            = fvdw;
 563
 564             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm_mul_ps(fscal,dx00);
 568             ty               = _mm_mul_ps(fscal,dy00);
 569             tz               = _mm_mul_ps(fscal,dz00);
 570
 571             /* Update vectorial force */
 572             fix0             = _mm_add_ps(fix0,tx);
 573             fiy0             = _mm_add_ps(fiy0,ty);
 574             fiz0             = _mm_add_ps(fiz0,tz);
 575
 576             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 577             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 578             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 579             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 580             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 581
 582             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             r10              = _mm_mul_ps(rsq10,rinv10);
 587             r10              = _mm_andnot_ps(dummy_mask,r10);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq10             = _mm_mul_ps(iq1,jq0);
 591
 592             /* Calculate table index by multiplying r with table scale and truncate to integer */
 593             rt               = _mm_mul_ps(r10,vftabscale);
 594             vfitab           = _mm_cvttps_epi32(rt);
 595             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 596             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 597
 598             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 599             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 600             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 601             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 602             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 603             _MM_TRANSPOSE4_PS(Y,F,G,H);
 604             Heps             = _mm_mul_ps(vfeps,H);
 605             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 606             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 607             velec            = _mm_mul_ps(qq10,VV);
 608             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 609             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 610
 611             /* Update potential sum for this i atom from the interaction with this j atom. */
 612             velec            = _mm_andnot_ps(dummy_mask,velec);
 613             velecsum         = _mm_add_ps(velecsum,velec);
 614
 615             fscal            = felec;
 616
 617             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm_mul_ps(fscal,dx10);
 621             ty               = _mm_mul_ps(fscal,dy10);
 622             tz               = _mm_mul_ps(fscal,dz10);
 623
 624             /* Update vectorial force */
 625             fix1             = _mm_add_ps(fix1,tx);
 626             fiy1             = _mm_add_ps(fiy1,ty);
 627             fiz1             = _mm_add_ps(fiz1,tz);
 628
 629             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 630             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 631             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 632             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 633             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 634
 635             /**************************
 636              * CALCULATE INTERACTIONS *
 637              **************************/
 638
 639             r20              = _mm_mul_ps(rsq20,rinv20);
 640             r20              = _mm_andnot_ps(dummy_mask,r20);
 641
 642             /* Compute parameters for interactions between i and j atoms */
 643             qq20             = _mm_mul_ps(iq2,jq0);
 644
 645             /* Calculate table index by multiplying r with table scale and truncate to integer */
 646             rt               = _mm_mul_ps(r20,vftabscale);
 647             vfitab           = _mm_cvttps_epi32(rt);
 648             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 649             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 650
 651             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 652             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 653             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 654             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 655             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 656             _MM_TRANSPOSE4_PS(Y,F,G,H);
 657             Heps             = _mm_mul_ps(vfeps,H);
 658             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 659             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 660             velec            = _mm_mul_ps(qq20,VV);
 661             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 662             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 663
 664             /* Update potential sum for this i atom from the interaction with this j atom. */
 665             velec            = _mm_andnot_ps(dummy_mask,velec);
 666             velecsum         = _mm_add_ps(velecsum,velec);
 667
 668             fscal            = felec;
 669
 670             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 671
 672             /* Calculate temporary vectorial force */
 673             tx               = _mm_mul_ps(fscal,dx20);
 674             ty               = _mm_mul_ps(fscal,dy20);
 675             tz               = _mm_mul_ps(fscal,dz20);
 676
 677             /* Update vectorial force */
 678             fix2             = _mm_add_ps(fix2,tx);
 679             fiy2             = _mm_add_ps(fiy2,ty);
 680             fiz2             = _mm_add_ps(fiz2,tz);
 681
 682             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 683             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 684             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 685             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 686             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 687
 688             /**************************
 689              * CALCULATE INTERACTIONS *
 690              **************************/
 691
 692             r30              = _mm_mul_ps(rsq30,rinv30);
 693             r30              = _mm_andnot_ps(dummy_mask,r30);
 694
 695             /* Compute parameters for interactions between i and j atoms */
 696             qq30             = _mm_mul_ps(iq3,jq0);
 697
 698             /* Calculate table index by multiplying r with table scale and truncate to integer */
 699             rt               = _mm_mul_ps(r30,vftabscale);
 700             vfitab           = _mm_cvttps_epi32(rt);
 701             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 702             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 703
 704             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 705             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 706             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 707             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 708             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 709             _MM_TRANSPOSE4_PS(Y,F,G,H);
 710             Heps             = _mm_mul_ps(vfeps,H);
 711             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 712             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 713             velec            = _mm_mul_ps(qq30,VV);
 714             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 715             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 716
 717             /* Update potential sum for this i atom from the interaction with this j atom. */
 718             velec            = _mm_andnot_ps(dummy_mask,velec);
 719             velecsum         = _mm_add_ps(velecsum,velec);
 720
 721             fscal            = felec;
 722
 723             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 724
 725             /* Calculate temporary vectorial force */
 726             tx               = _mm_mul_ps(fscal,dx30);
 727             ty               = _mm_mul_ps(fscal,dy30);
 728             tz               = _mm_mul_ps(fscal,dz30);
 729
 730             /* Update vectorial force */
 731             fix3             = _mm_add_ps(fix3,tx);
 732             fiy3             = _mm_add_ps(fiy3,ty);
 733             fiz3             = _mm_add_ps(fiz3,tz);
 734
 735             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 736             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 737             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 738             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 739             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 740
 741             /* Inner loop uses 189 flops */
 742         }
 743
 744         /* End of innermost loop */
 745
 746         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 747                                               f+i_coord_offset,fshift+i_shift_offset);
 748
 749         ggid                        = gid[iidx];
 750         /* Update potential energies */
 751         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 752         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 753
 754         /* Increment number of inner iterations */
 755         inneriter                  += j_index_end - j_index_start;
 756
 757         /* Outer loop uses 26 flops */
 758     }
 759
 760     /* Increment number of outer iterations */
 761     outeriter        += nri;
 762
 763     /* Update outer/inner flops */
 764
 765     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*189);
 766 }
 767 /*
 768  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse2_single
 769  * Electrostatics interaction: CubicSplineTable
 770  * VdW interaction:            CubicSplineTable
 771  * Geometry:                   Water4-Particle
 772  * Calculate force/pot:        Force
 773  */
 774 void
 775 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse2_single
 776                     (t_nblist * gmx_restrict                nlist,
 777                      rvec * gmx_restrict                    xx,
 778                      rvec * gmx_restrict                    ff,
 779                      t_forcerec * gmx_restrict              fr,
 780                      t_mdatoms * gmx_restrict               mdatoms,
 781                      nb_kernel_data_t * gmx_restrict        kernel_data,
 782                      t_nrnb * gmx_restrict                  nrnb)
 783 {
 784     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 785      * just 0 for non-waters.
 786      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 787      * jnr indices corresponding to data put in the four positions in the SIMD register.
 788      */
 789     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 790     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 791     int              jnrA,jnrB,jnrC,jnrD;
 792     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 793     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 794     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 795     real             rcutoff_scalar;
 796     real             *shiftvec,*fshift,*x,*f;
 797     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 798     real             scratch[4*DIM];
 799     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 800     int              vdwioffset0;
 801     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 802     int              vdwioffset1;
 803     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 804     int              vdwioffset2;
 805     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 806     int              vdwioffset3;
 807     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 808     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 809     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 810     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 811     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 812     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 813     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 814     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 815     real             *charge;
 816     int              nvdwtype;
 817     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 818     int              *vdwtype;
 819     real             *vdwparam;
 820     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 821     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 822     __m128i          vfitab;
 823     __m128i          ifour       = _mm_set1_epi32(4);
 824     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 825     real             *vftab;
 826     __m128           dummy_mask,cutoff_mask;
 827     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 828     __m128           one     = _mm_set1_ps(1.0);
 829     __m128           two     = _mm_set1_ps(2.0);
 830     x                = xx[0];
 831     f                = ff[0];
 832
 833     nri              = nlist->nri;
 834     iinr             = nlist->iinr;
 835     jindex           = nlist->jindex;
 836     jjnr             = nlist->jjnr;
 837     shiftidx         = nlist->shift;
 838     gid              = nlist->gid;
 839     shiftvec         = fr->shift_vec[0];
 840     fshift           = fr->fshift[0];
 841     facel            = _mm_set1_ps(fr->epsfac);
 842     charge           = mdatoms->chargeA;
 843     nvdwtype         = fr->ntype;
 844     vdwparam         = fr->nbfp;
 845     vdwtype          = mdatoms->typeA;
 846
 847     vftab            = kernel_data->table_elec_vdw->data;
 848     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 849
 850     /* Setup water-specific parameters */
 851     inr              = nlist->iinr[0];
 852     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 853     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 854     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 855     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 856
 857     /* Avoid stupid compiler warnings */
 858     jnrA = jnrB = jnrC = jnrD = 0;
 859     j_coord_offsetA = 0;
 860     j_coord_offsetB = 0;
 861     j_coord_offsetC = 0;
 862     j_coord_offsetD = 0;
 863
 864     outeriter        = 0;
 865     inneriter        = 0;
 866
 867     for(iidx=0;iidx<4*DIM;iidx++)
 868     {
 869         scratch[iidx] = 0.0;
 870     }
 871
 872     /* Start outer loop over neighborlists */
 873     for(iidx=0; iidx<nri; iidx++)
 874     {
 875         /* Load shift vector for this list */
 876         i_shift_offset   = DIM*shiftidx[iidx];
 877
 878         /* Load limits for loop over neighbors */
 879         j_index_start    = jindex[iidx];
 880         j_index_end      = jindex[iidx+1];
 881
 882         /* Get outer coordinate index */
 883         inr              = iinr[iidx];
 884         i_coord_offset   = DIM*inr;
 885
 886         /* Load i particle coords and add shift vector */
 887         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 888                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 889
 890         fix0             = _mm_setzero_ps();
 891         fiy0             = _mm_setzero_ps();
 892         fiz0             = _mm_setzero_ps();
 893         fix1             = _mm_setzero_ps();
 894         fiy1             = _mm_setzero_ps();
 895         fiz1             = _mm_setzero_ps();
 896         fix2             = _mm_setzero_ps();
 897         fiy2             = _mm_setzero_ps();
 898         fiz2             = _mm_setzero_ps();
 899         fix3             = _mm_setzero_ps();
 900         fiy3             = _mm_setzero_ps();
 901         fiz3             = _mm_setzero_ps();
 902
 903         /* Start inner kernel loop */
 904         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 905         {
 906
 907             /* Get j neighbor index, and coordinate index */
 908             jnrA             = jjnr[jidx];
 909             jnrB             = jjnr[jidx+1];
 910             jnrC             = jjnr[jidx+2];
 911             jnrD             = jjnr[jidx+3];
 912             j_coord_offsetA  = DIM*jnrA;
 913             j_coord_offsetB  = DIM*jnrB;
 914             j_coord_offsetC  = DIM*jnrC;
 915             j_coord_offsetD  = DIM*jnrD;
 916
 917             /* load j atom coordinates */
 918             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 919                                               x+j_coord_offsetC,x+j_coord_offsetD,
 920                                               &jx0,&jy0,&jz0);
 921
 922             /* Calculate displacement vector */
 923             dx00             = _mm_sub_ps(ix0,jx0);
 924             dy00             = _mm_sub_ps(iy0,jy0);
 925             dz00             = _mm_sub_ps(iz0,jz0);
 926             dx10             = _mm_sub_ps(ix1,jx0);
 927             dy10             = _mm_sub_ps(iy1,jy0);
 928             dz10             = _mm_sub_ps(iz1,jz0);
 929             dx20             = _mm_sub_ps(ix2,jx0);
 930             dy20             = _mm_sub_ps(iy2,jy0);
 931             dz20             = _mm_sub_ps(iz2,jz0);
 932             dx30             = _mm_sub_ps(ix3,jx0);
 933             dy30             = _mm_sub_ps(iy3,jy0);
 934             dz30             = _mm_sub_ps(iz3,jz0);
 935
 936             /* Calculate squared distance and things based on it */
 937             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 938             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 939             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 940             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 941
 942             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 943             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 944             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 945             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 946
 947             /* Load parameters for j particles */
 948             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 949                                                               charge+jnrC+0,charge+jnrD+0);
 950             vdwjidx0A        = 2*vdwtype[jnrA+0];
 951             vdwjidx0B        = 2*vdwtype[jnrB+0];
 952             vdwjidx0C        = 2*vdwtype[jnrC+0];
 953             vdwjidx0D        = 2*vdwtype[jnrD+0];
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             r00              = _mm_mul_ps(rsq00,rinv00);
 960
 961             /* Compute parameters for interactions between i and j atoms */
 962             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 963                                          vdwparam+vdwioffset0+vdwjidx0B,
 964                                          vdwparam+vdwioffset0+vdwjidx0C,
 965                                          vdwparam+vdwioffset0+vdwjidx0D,
 966                                          &c6_00,&c12_00);
 967
 968             /* Calculate table index by multiplying r with table scale and truncate to integer */
 969             rt               = _mm_mul_ps(r00,vftabscale);
 970             vfitab           = _mm_cvttps_epi32(rt);
 971             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 972             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 973
 974             /* CUBIC SPLINE TABLE DISPERSION */
 975             vfitab           = _mm_add_epi32(vfitab,ifour);
 976             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 977             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 978             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 979             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 980             _MM_TRANSPOSE4_PS(Y,F,G,H);
 981             Heps             = _mm_mul_ps(vfeps,H);
 982             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 983             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 984             fvdw6            = _mm_mul_ps(c6_00,FF);
 985
 986             /* CUBIC SPLINE TABLE REPULSION */
 987             vfitab           = _mm_add_epi32(vfitab,ifour);
 988             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 989             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 990             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 991             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 992             _MM_TRANSPOSE4_PS(Y,F,G,H);
 993             Heps             = _mm_mul_ps(vfeps,H);
 994             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 995             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 996             fvdw12           = _mm_mul_ps(c12_00,FF);
 997             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 998
 999             fscal            = fvdw;
1000
1001             /* Calculate temporary vectorial force */
1002             tx               = _mm_mul_ps(fscal,dx00);
1003             ty               = _mm_mul_ps(fscal,dy00);
1004             tz               = _mm_mul_ps(fscal,dz00);
1005
1006             /* Update vectorial force */
1007             fix0             = _mm_add_ps(fix0,tx);
1008             fiy0             = _mm_add_ps(fiy0,ty);
1009             fiz0             = _mm_add_ps(fiz0,tz);
1010
1011             fjptrA             = f+j_coord_offsetA;
1012             fjptrB             = f+j_coord_offsetB;
1013             fjptrC             = f+j_coord_offsetC;
1014             fjptrD             = f+j_coord_offsetD;
1015             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1016
1017             /**************************
1018              * CALCULATE INTERACTIONS *
1019              **************************/
1020
1021             r10              = _mm_mul_ps(rsq10,rinv10);
1022
1023             /* Compute parameters for interactions between i and j atoms */
1024             qq10             = _mm_mul_ps(iq1,jq0);
1025
1026             /* Calculate table index by multiplying r with table scale and truncate to integer */
1027             rt               = _mm_mul_ps(r10,vftabscale);
1028             vfitab           = _mm_cvttps_epi32(rt);
1029             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1030             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1031
1032             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1033             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1034             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1035             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1036             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1037             _MM_TRANSPOSE4_PS(Y,F,G,H);
1038             Heps             = _mm_mul_ps(vfeps,H);
1039             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1040             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1041             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1042
1043             fscal            = felec;
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm_mul_ps(fscal,dx10);
1047             ty               = _mm_mul_ps(fscal,dy10);
1048             tz               = _mm_mul_ps(fscal,dz10);
1049
1050             /* Update vectorial force */
1051             fix1             = _mm_add_ps(fix1,tx);
1052             fiy1             = _mm_add_ps(fiy1,ty);
1053             fiz1             = _mm_add_ps(fiz1,tz);
1054
1055             fjptrA             = f+j_coord_offsetA;
1056             fjptrB             = f+j_coord_offsetB;
1057             fjptrC             = f+j_coord_offsetC;
1058             fjptrD             = f+j_coord_offsetD;
1059             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1060
1061             /**************************
1062              * CALCULATE INTERACTIONS *
1063              **************************/
1064
1065             r20              = _mm_mul_ps(rsq20,rinv20);
1066
1067             /* Compute parameters for interactions between i and j atoms */
1068             qq20             = _mm_mul_ps(iq2,jq0);
1069
1070             /* Calculate table index by multiplying r with table scale and truncate to integer */
1071             rt               = _mm_mul_ps(r20,vftabscale);
1072             vfitab           = _mm_cvttps_epi32(rt);
1073             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1074             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1075
1076             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1077             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1078             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1079             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1080             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1081             _MM_TRANSPOSE4_PS(Y,F,G,H);
1082             Heps             = _mm_mul_ps(vfeps,H);
1083             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1084             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1085             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1086
1087             fscal            = felec;
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm_mul_ps(fscal,dx20);
1091             ty               = _mm_mul_ps(fscal,dy20);
1092             tz               = _mm_mul_ps(fscal,dz20);
1093
1094             /* Update vectorial force */
1095             fix2             = _mm_add_ps(fix2,tx);
1096             fiy2             = _mm_add_ps(fiy2,ty);
1097             fiz2             = _mm_add_ps(fiz2,tz);
1098
1099             fjptrA             = f+j_coord_offsetA;
1100             fjptrB             = f+j_coord_offsetB;
1101             fjptrC             = f+j_coord_offsetC;
1102             fjptrD             = f+j_coord_offsetD;
1103             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1104
1105             /**************************
1106              * CALCULATE INTERACTIONS *
1107              **************************/
1108
1109             r30              = _mm_mul_ps(rsq30,rinv30);
1110
1111             /* Compute parameters for interactions between i and j atoms */
1112             qq30             = _mm_mul_ps(iq3,jq0);
1113
1114             /* Calculate table index by multiplying r with table scale and truncate to integer */
1115             rt               = _mm_mul_ps(r30,vftabscale);
1116             vfitab           = _mm_cvttps_epi32(rt);
1117             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1118             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1119
1120             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1121             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1122             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1123             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1124             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1125             _MM_TRANSPOSE4_PS(Y,F,G,H);
1126             Heps             = _mm_mul_ps(vfeps,H);
1127             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1128             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1129             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1130
1131             fscal            = felec;
1132
1133             /* Calculate temporary vectorial force */
1134             tx               = _mm_mul_ps(fscal,dx30);
1135             ty               = _mm_mul_ps(fscal,dy30);
1136             tz               = _mm_mul_ps(fscal,dz30);
1137
1138             /* Update vectorial force */
1139             fix3             = _mm_add_ps(fix3,tx);
1140             fiy3             = _mm_add_ps(fiy3,ty);
1141             fiz3             = _mm_add_ps(fiz3,tz);
1142
1143             fjptrA             = f+j_coord_offsetA;
1144             fjptrB             = f+j_coord_offsetB;
1145             fjptrC             = f+j_coord_offsetC;
1146             fjptrD             = f+j_coord_offsetD;
1147             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1148
1149             /* Inner loop uses 165 flops */
1150         }
1151
1152         if(jidx<j_index_end)
1153         {
1154
1155             /* Get j neighbor index, and coordinate index */
1156             jnrlistA         = jjnr[jidx];
1157             jnrlistB         = jjnr[jidx+1];
1158             jnrlistC         = jjnr[jidx+2];
1159             jnrlistD         = jjnr[jidx+3];
1160             /* Sign of each element will be negative for non-real atoms.
1161              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1162              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1163              */
1164             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1165             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1166             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1167             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1168             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1169             j_coord_offsetA  = DIM*jnrA;
1170             j_coord_offsetB  = DIM*jnrB;
1171             j_coord_offsetC  = DIM*jnrC;
1172             j_coord_offsetD  = DIM*jnrD;
1173
1174             /* load j atom coordinates */
1175             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1176                                               x+j_coord_offsetC,x+j_coord_offsetD,
1177                                               &jx0,&jy0,&jz0);
1178
1179             /* Calculate displacement vector */
1180             dx00             = _mm_sub_ps(ix0,jx0);
1181             dy00             = _mm_sub_ps(iy0,jy0);
1182             dz00             = _mm_sub_ps(iz0,jz0);
1183             dx10             = _mm_sub_ps(ix1,jx0);
1184             dy10             = _mm_sub_ps(iy1,jy0);
1185             dz10             = _mm_sub_ps(iz1,jz0);
1186             dx20             = _mm_sub_ps(ix2,jx0);
1187             dy20             = _mm_sub_ps(iy2,jy0);
1188             dz20             = _mm_sub_ps(iz2,jz0);
1189             dx30             = _mm_sub_ps(ix3,jx0);
1190             dy30             = _mm_sub_ps(iy3,jy0);
1191             dz30             = _mm_sub_ps(iz3,jz0);
1192
1193             /* Calculate squared distance and things based on it */
1194             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1195             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1196             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1197             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1198
1199             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1200             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1201             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1202             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1203
1204             /* Load parameters for j particles */
1205             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1206                                                               charge+jnrC+0,charge+jnrD+0);
1207             vdwjidx0A        = 2*vdwtype[jnrA+0];
1208             vdwjidx0B        = 2*vdwtype[jnrB+0];
1209             vdwjidx0C        = 2*vdwtype[jnrC+0];
1210             vdwjidx0D        = 2*vdwtype[jnrD+0];
1211
1212             /**************************
1213              * CALCULATE INTERACTIONS *
1214              **************************/
1215
1216             r00              = _mm_mul_ps(rsq00,rinv00);
1217             r00              = _mm_andnot_ps(dummy_mask,r00);
1218
1219             /* Compute parameters for interactions between i and j atoms */
1220             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1221                                          vdwparam+vdwioffset0+vdwjidx0B,
1222                                          vdwparam+vdwioffset0+vdwjidx0C,
1223                                          vdwparam+vdwioffset0+vdwjidx0D,
1224                                          &c6_00,&c12_00);
1225
1226             /* Calculate table index by multiplying r with table scale and truncate to integer */
1227             rt               = _mm_mul_ps(r00,vftabscale);
1228             vfitab           = _mm_cvttps_epi32(rt);
1229             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1230             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1231
1232             /* CUBIC SPLINE TABLE DISPERSION */
1233             vfitab           = _mm_add_epi32(vfitab,ifour);
1234             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1235             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1236             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1237             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1238             _MM_TRANSPOSE4_PS(Y,F,G,H);
1239             Heps             = _mm_mul_ps(vfeps,H);
1240             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1241             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1242             fvdw6            = _mm_mul_ps(c6_00,FF);
1243
1244             /* CUBIC SPLINE TABLE REPULSION */
1245             vfitab           = _mm_add_epi32(vfitab,ifour);
1246             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1247             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1248             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1249             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1250             _MM_TRANSPOSE4_PS(Y,F,G,H);
1251             Heps             = _mm_mul_ps(vfeps,H);
1252             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1253             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1254             fvdw12           = _mm_mul_ps(c12_00,FF);
1255             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1256
1257             fscal            = fvdw;
1258
1259             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1260
1261             /* Calculate temporary vectorial force */
1262             tx               = _mm_mul_ps(fscal,dx00);
1263             ty               = _mm_mul_ps(fscal,dy00);
1264             tz               = _mm_mul_ps(fscal,dz00);
1265
1266             /* Update vectorial force */
1267             fix0             = _mm_add_ps(fix0,tx);
1268             fiy0             = _mm_add_ps(fiy0,ty);
1269             fiz0             = _mm_add_ps(fiz0,tz);
1270
1271             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1272             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1273             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1274             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1275             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1276
1277             /**************************
1278              * CALCULATE INTERACTIONS *
1279              **************************/
1280
1281             r10              = _mm_mul_ps(rsq10,rinv10);
1282             r10              = _mm_andnot_ps(dummy_mask,r10);
1283
1284             /* Compute parameters for interactions between i and j atoms */
1285             qq10             = _mm_mul_ps(iq1,jq0);
1286
1287             /* Calculate table index by multiplying r with table scale and truncate to integer */
1288             rt               = _mm_mul_ps(r10,vftabscale);
1289             vfitab           = _mm_cvttps_epi32(rt);
1290             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1291             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1292
1293             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1294             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1295             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1296             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1297             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1298             _MM_TRANSPOSE4_PS(Y,F,G,H);
1299             Heps             = _mm_mul_ps(vfeps,H);
1300             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1301             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1302             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1303
1304             fscal            = felec;
1305
1306             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1307
1308             /* Calculate temporary vectorial force */
1309             tx               = _mm_mul_ps(fscal,dx10);
1310             ty               = _mm_mul_ps(fscal,dy10);
1311             tz               = _mm_mul_ps(fscal,dz10);
1312
1313             /* Update vectorial force */
1314             fix1             = _mm_add_ps(fix1,tx);
1315             fiy1             = _mm_add_ps(fiy1,ty);
1316             fiz1             = _mm_add_ps(fiz1,tz);
1317
1318             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1319             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1320             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1321             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1322             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1323
1324             /**************************
1325              * CALCULATE INTERACTIONS *
1326              **************************/
1327
1328             r20              = _mm_mul_ps(rsq20,rinv20);
1329             r20              = _mm_andnot_ps(dummy_mask,r20);
1330
1331             /* Compute parameters for interactions between i and j atoms */
1332             qq20             = _mm_mul_ps(iq2,jq0);
1333
1334             /* Calculate table index by multiplying r with table scale and truncate to integer */
1335             rt               = _mm_mul_ps(r20,vftabscale);
1336             vfitab           = _mm_cvttps_epi32(rt);
1337             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1338             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1339
1340             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1341             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1342             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1343             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1344             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1345             _MM_TRANSPOSE4_PS(Y,F,G,H);
1346             Heps             = _mm_mul_ps(vfeps,H);
1347             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1348             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1349             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1350
1351             fscal            = felec;
1352
1353             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1354
1355             /* Calculate temporary vectorial force */
1356             tx               = _mm_mul_ps(fscal,dx20);
1357             ty               = _mm_mul_ps(fscal,dy20);
1358             tz               = _mm_mul_ps(fscal,dz20);
1359
1360             /* Update vectorial force */
1361             fix2             = _mm_add_ps(fix2,tx);
1362             fiy2             = _mm_add_ps(fiy2,ty);
1363             fiz2             = _mm_add_ps(fiz2,tz);
1364
1365             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1366             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1367             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1368             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1369             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1370
1371             /**************************
1372              * CALCULATE INTERACTIONS *
1373              **************************/
1374
1375             r30              = _mm_mul_ps(rsq30,rinv30);
1376             r30              = _mm_andnot_ps(dummy_mask,r30);
1377
1378             /* Compute parameters for interactions between i and j atoms */
1379             qq30             = _mm_mul_ps(iq3,jq0);
1380
1381             /* Calculate table index by multiplying r with table scale and truncate to integer */
1382             rt               = _mm_mul_ps(r30,vftabscale);
1383             vfitab           = _mm_cvttps_epi32(rt);
1384             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1385             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1386
1387             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1388             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1389             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1390             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1391             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1392             _MM_TRANSPOSE4_PS(Y,F,G,H);
1393             Heps             = _mm_mul_ps(vfeps,H);
1394             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1395             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1396             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1397
1398             fscal            = felec;
1399
1400             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1401
1402             /* Calculate temporary vectorial force */
1403             tx               = _mm_mul_ps(fscal,dx30);
1404             ty               = _mm_mul_ps(fscal,dy30);
1405             tz               = _mm_mul_ps(fscal,dz30);
1406
1407             /* Update vectorial force */
1408             fix3             = _mm_add_ps(fix3,tx);
1409             fiy3             = _mm_add_ps(fiy3,ty);
1410             fiz3             = _mm_add_ps(fiz3,tz);
1411
1412             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1413             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1414             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1415             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1416             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1417
1418             /* Inner loop uses 169 flops */
1419         }
1420
1421         /* End of innermost loop */
1422
1423         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1424                                               f+i_coord_offset,fshift+i_shift_offset);
1425
1426         /* Increment number of inner iterations */
1427         inneriter                  += j_index_end - j_index_start;
1428
1429         /* Outer loop uses 24 flops */
1430     }
1431
1432     /* Increment number of outer iterations */
1433     outeriter        += nri;
1434
1435     /* Update outer/inner flops */
1436
1437     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*169);
1438 }