src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwLJ_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_VdwLJ_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_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->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_elec->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             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 216             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 217             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 218
 219             rinvsq00         = gmx_mm_inv_ps(rsq00);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 223                                                               charge+jnrC+0,charge+jnrD+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             /* Compute parameters for interactions between i and j atoms */
 234             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 235                                          vdwparam+vdwioffset0+vdwjidx0B,
 236                                          vdwparam+vdwioffset0+vdwjidx0C,
 237                                          vdwparam+vdwioffset0+vdwjidx0D,
 238                                          &c6_00,&c12_00);
 239
 240             /* LENNARD-JONES DISPERSION/REPULSION */
 241
 242             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 243             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 244             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 245             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 246             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 250
 251             fscal            = fvdw;
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_ps(fscal,dx00);
 255             ty               = _mm_mul_ps(fscal,dy00);
 256             tz               = _mm_mul_ps(fscal,dz00);
 257
 258             /* Update vectorial force */
 259             fix0             = _mm_add_ps(fix0,tx);
 260             fiy0             = _mm_add_ps(fiy0,ty);
 261             fiz0             = _mm_add_ps(fiz0,tz);
 262
 263             fjptrA             = f+j_coord_offsetA;
 264             fjptrB             = f+j_coord_offsetB;
 265             fjptrC             = f+j_coord_offsetC;
 266             fjptrD             = f+j_coord_offsetD;
 267             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             r10              = _mm_mul_ps(rsq10,rinv10);
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             qq10             = _mm_mul_ps(iq1,jq0);
 277
 278             /* Calculate table index by multiplying r with table scale and truncate to integer */
 279             rt               = _mm_mul_ps(r10,vftabscale);
 280             vfitab           = _mm_cvttps_epi32(rt);
 281             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 282             vfitab           = _mm_slli_epi32(vfitab,2);
 283
 284             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 285             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 286             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 287             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 288             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 289             _MM_TRANSPOSE4_PS(Y,F,G,H);
 290             Heps             = _mm_mul_ps(vfeps,H);
 291             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 292             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 293             velec            = _mm_mul_ps(qq10,VV);
 294             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 295             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm_mul_ps(fscal,dx10);
 304             ty               = _mm_mul_ps(fscal,dy10);
 305             tz               = _mm_mul_ps(fscal,dz10);
 306
 307             /* Update vectorial force */
 308             fix1             = _mm_add_ps(fix1,tx);
 309             fiy1             = _mm_add_ps(fiy1,ty);
 310             fiz1             = _mm_add_ps(fiz1,tz);
 311
 312             fjptrA             = f+j_coord_offsetA;
 313             fjptrB             = f+j_coord_offsetB;
 314             fjptrC             = f+j_coord_offsetC;
 315             fjptrD             = f+j_coord_offsetD;
 316             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             r20              = _mm_mul_ps(rsq20,rinv20);
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq20             = _mm_mul_ps(iq2,jq0);
 326
 327             /* Calculate table index by multiplying r with table scale and truncate to integer */
 328             rt               = _mm_mul_ps(r20,vftabscale);
 329             vfitab           = _mm_cvttps_epi32(rt);
 330             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 331             vfitab           = _mm_slli_epi32(vfitab,2);
 332
 333             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 334             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 335             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 336             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 337             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 338             _MM_TRANSPOSE4_PS(Y,F,G,H);
 339             Heps             = _mm_mul_ps(vfeps,H);
 340             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 341             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 342             velec            = _mm_mul_ps(qq20,VV);
 343             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 344             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm_mul_ps(fscal,dx20);
 353             ty               = _mm_mul_ps(fscal,dy20);
 354             tz               = _mm_mul_ps(fscal,dz20);
 355
 356             /* Update vectorial force */
 357             fix2             = _mm_add_ps(fix2,tx);
 358             fiy2             = _mm_add_ps(fiy2,ty);
 359             fiz2             = _mm_add_ps(fiz2,tz);
 360
 361             fjptrA             = f+j_coord_offsetA;
 362             fjptrB             = f+j_coord_offsetB;
 363             fjptrC             = f+j_coord_offsetC;
 364             fjptrD             = f+j_coord_offsetD;
 365             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             r30              = _mm_mul_ps(rsq30,rinv30);
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq30             = _mm_mul_ps(iq3,jq0);
 375
 376             /* Calculate table index by multiplying r with table scale and truncate to integer */
 377             rt               = _mm_mul_ps(r30,vftabscale);
 378             vfitab           = _mm_cvttps_epi32(rt);
 379             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 380             vfitab           = _mm_slli_epi32(vfitab,2);
 381
 382             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 383             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 384             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 385             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 386             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 387             _MM_TRANSPOSE4_PS(Y,F,G,H);
 388             Heps             = _mm_mul_ps(vfeps,H);
 389             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 390             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 391             velec            = _mm_mul_ps(qq30,VV);
 392             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 393             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velecsum         = _mm_add_ps(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             /* Calculate temporary vectorial force */
 401             tx               = _mm_mul_ps(fscal,dx30);
 402             ty               = _mm_mul_ps(fscal,dy30);
 403             tz               = _mm_mul_ps(fscal,dz30);
 404
 405             /* Update vectorial force */
 406             fix3             = _mm_add_ps(fix3,tx);
 407             fiy3             = _mm_add_ps(fiy3,ty);
 408             fiz3             = _mm_add_ps(fiz3,tz);
 409
 410             fjptrA             = f+j_coord_offsetA;
 411             fjptrB             = f+j_coord_offsetB;
 412             fjptrC             = f+j_coord_offsetC;
 413             fjptrD             = f+j_coord_offsetD;
 414             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 415
 416             /* Inner loop uses 161 flops */
 417         }
 418
 419         if(jidx<j_index_end)
 420         {
 421
 422             /* Get j neighbor index, and coordinate index */
 423             jnrlistA         = jjnr[jidx];
 424             jnrlistB         = jjnr[jidx+1];
 425             jnrlistC         = jjnr[jidx+2];
 426             jnrlistD         = jjnr[jidx+3];
 427             /* Sign of each element will be negative for non-real atoms.
 428              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 429              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 430              */
 431             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 432             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 433             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 434             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 435             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438             j_coord_offsetC  = DIM*jnrC;
 439             j_coord_offsetD  = DIM*jnrD;
 440
 441             /* load j atom coordinates */
 442             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 443                                               x+j_coord_offsetC,x+j_coord_offsetD,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm_sub_ps(ix0,jx0);
 448             dy00             = _mm_sub_ps(iy0,jy0);
 449             dz00             = _mm_sub_ps(iz0,jz0);
 450             dx10             = _mm_sub_ps(ix1,jx0);
 451             dy10             = _mm_sub_ps(iy1,jy0);
 452             dz10             = _mm_sub_ps(iz1,jz0);
 453             dx20             = _mm_sub_ps(ix2,jx0);
 454             dy20             = _mm_sub_ps(iy2,jy0);
 455             dz20             = _mm_sub_ps(iz2,jz0);
 456             dx30             = _mm_sub_ps(ix3,jx0);
 457             dy30             = _mm_sub_ps(iy3,jy0);
 458             dz30             = _mm_sub_ps(iz3,jz0);
 459
 460             /* Calculate squared distance and things based on it */
 461             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 462             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 463             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 464             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 465
 466             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 467             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 468             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 469
 470             rinvsq00         = gmx_mm_inv_ps(rsq00);
 471
 472             /* Load parameters for j particles */
 473             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 474                                                               charge+jnrC+0,charge+jnrD+0);
 475             vdwjidx0A        = 2*vdwtype[jnrA+0];
 476             vdwjidx0B        = 2*vdwtype[jnrB+0];
 477             vdwjidx0C        = 2*vdwtype[jnrC+0];
 478             vdwjidx0D        = 2*vdwtype[jnrD+0];
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 486                                          vdwparam+vdwioffset0+vdwjidx0B,
 487                                          vdwparam+vdwioffset0+vdwjidx0C,
 488                                          vdwparam+vdwioffset0+vdwjidx0D,
 489                                          &c6_00,&c12_00);
 490
 491             /* LENNARD-JONES DISPERSION/REPULSION */
 492
 493             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 494             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 495             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 496             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 497             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 498
 499             /* Update potential sum for this i atom from the interaction with this j atom. */
 500             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 501             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 502
 503             fscal            = fvdw;
 504
 505             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 506
 507             /* Calculate temporary vectorial force */
 508             tx               = _mm_mul_ps(fscal,dx00);
 509             ty               = _mm_mul_ps(fscal,dy00);
 510             tz               = _mm_mul_ps(fscal,dz00);
 511
 512             /* Update vectorial force */
 513             fix0             = _mm_add_ps(fix0,tx);
 514             fiy0             = _mm_add_ps(fiy0,ty);
 515             fiz0             = _mm_add_ps(fiz0,tz);
 516
 517             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 518             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 519             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 520             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 521             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             r10              = _mm_mul_ps(rsq10,rinv10);
 528             r10              = _mm_andnot_ps(dummy_mask,r10);
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq10             = _mm_mul_ps(iq1,jq0);
 532
 533             /* Calculate table index by multiplying r with table scale and truncate to integer */
 534             rt               = _mm_mul_ps(r10,vftabscale);
 535             vfitab           = _mm_cvttps_epi32(rt);
 536             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 537             vfitab           = _mm_slli_epi32(vfitab,2);
 538
 539             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 540             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 541             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 542             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 543             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 544             _MM_TRANSPOSE4_PS(Y,F,G,H);
 545             Heps             = _mm_mul_ps(vfeps,H);
 546             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 547             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 548             velec            = _mm_mul_ps(qq10,VV);
 549             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 550             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm_andnot_ps(dummy_mask,velec);
 554             velecsum         = _mm_add_ps(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 559
 560             /* Calculate temporary vectorial force */
 561             tx               = _mm_mul_ps(fscal,dx10);
 562             ty               = _mm_mul_ps(fscal,dy10);
 563             tz               = _mm_mul_ps(fscal,dz10);
 564
 565             /* Update vectorial force */
 566             fix1             = _mm_add_ps(fix1,tx);
 567             fiy1             = _mm_add_ps(fiy1,ty);
 568             fiz1             = _mm_add_ps(fiz1,tz);
 569
 570             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 571             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 572             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 573             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 574             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 575
 576             /**************************
 577              * CALCULATE INTERACTIONS *
 578              **************************/
 579
 580             r20              = _mm_mul_ps(rsq20,rinv20);
 581             r20              = _mm_andnot_ps(dummy_mask,r20);
 582
 583             /* Compute parameters for interactions between i and j atoms */
 584             qq20             = _mm_mul_ps(iq2,jq0);
 585
 586             /* Calculate table index by multiplying r with table scale and truncate to integer */
 587             rt               = _mm_mul_ps(r20,vftabscale);
 588             vfitab           = _mm_cvttps_epi32(rt);
 589             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 590             vfitab           = _mm_slli_epi32(vfitab,2);
 591
 592             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 593             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 594             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 595             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 596             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 597             _MM_TRANSPOSE4_PS(Y,F,G,H);
 598             Heps             = _mm_mul_ps(vfeps,H);
 599             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 600             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 601             velec            = _mm_mul_ps(qq20,VV);
 602             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 603             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm_andnot_ps(dummy_mask,velec);
 607             velecsum         = _mm_add_ps(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm_mul_ps(fscal,dx20);
 615             ty               = _mm_mul_ps(fscal,dy20);
 616             tz               = _mm_mul_ps(fscal,dz20);
 617
 618             /* Update vectorial force */
 619             fix2             = _mm_add_ps(fix2,tx);
 620             fiy2             = _mm_add_ps(fiy2,ty);
 621             fiz2             = _mm_add_ps(fiz2,tz);
 622
 623             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 624             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 625             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 626             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 627             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             r30              = _mm_mul_ps(rsq30,rinv30);
 634             r30              = _mm_andnot_ps(dummy_mask,r30);
 635
 636             /* Compute parameters for interactions between i and j atoms */
 637             qq30             = _mm_mul_ps(iq3,jq0);
 638
 639             /* Calculate table index by multiplying r with table scale and truncate to integer */
 640             rt               = _mm_mul_ps(r30,vftabscale);
 641             vfitab           = _mm_cvttps_epi32(rt);
 642             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 643             vfitab           = _mm_slli_epi32(vfitab,2);
 644
 645             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 646             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 647             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 648             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 649             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 650             _MM_TRANSPOSE4_PS(Y,F,G,H);
 651             Heps             = _mm_mul_ps(vfeps,H);
 652             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 653             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 654             velec            = _mm_mul_ps(qq30,VV);
 655             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 656             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 657
 658             /* Update potential sum for this i atom from the interaction with this j atom. */
 659             velec            = _mm_andnot_ps(dummy_mask,velec);
 660             velecsum         = _mm_add_ps(velecsum,velec);
 661
 662             fscal            = felec;
 663
 664             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 665
 666             /* Calculate temporary vectorial force */
 667             tx               = _mm_mul_ps(fscal,dx30);
 668             ty               = _mm_mul_ps(fscal,dy30);
 669             tz               = _mm_mul_ps(fscal,dz30);
 670
 671             /* Update vectorial force */
 672             fix3             = _mm_add_ps(fix3,tx);
 673             fiy3             = _mm_add_ps(fiy3,ty);
 674             fiz3             = _mm_add_ps(fiz3,tz);
 675
 676             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 677             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 678             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 679             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 680             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 681
 682             /* Inner loop uses 164 flops */
 683         }
 684
 685         /* End of innermost loop */
 686
 687         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 688                                               f+i_coord_offset,fshift+i_shift_offset);
 689
 690         ggid                        = gid[iidx];
 691         /* Update potential energies */
 692         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 693         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 694
 695         /* Increment number of inner iterations */
 696         inneriter                  += j_index_end - j_index_start;
 697
 698         /* Outer loop uses 26 flops */
 699     }
 700
 701     /* Increment number of outer iterations */
 702     outeriter        += nri;
 703
 704     /* Update outer/inner flops */
 705
 706     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*164);
 707 }
 708 /*
 709  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_single
 710  * Electrostatics interaction: CubicSplineTable
 711  * VdW interaction:            LennardJones
 712  * Geometry:                   Water4-Particle
 713  * Calculate force/pot:        Force
 714  */
 715 void
 716 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_single
 717                     (t_nblist * gmx_restrict                nlist,
 718                      rvec * gmx_restrict                    xx,
 719                      rvec * gmx_restrict                    ff,
 720                      t_forcerec * gmx_restrict              fr,
 721                      t_mdatoms * gmx_restrict               mdatoms,
 722                      nb_kernel_data_t * gmx_restrict        kernel_data,
 723                      t_nrnb * gmx_restrict                  nrnb)
 724 {
 725     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 726      * just 0 for non-waters.
 727      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 728      * jnr indices corresponding to data put in the four positions in the SIMD register.
 729      */
 730     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 731     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 732     int              jnrA,jnrB,jnrC,jnrD;
 733     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 734     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 735     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 736     real             rcutoff_scalar;
 737     real             *shiftvec,*fshift,*x,*f;
 738     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 739     real             scratch[4*DIM];
 740     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 741     int              vdwioffset0;
 742     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 743     int              vdwioffset1;
 744     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 745     int              vdwioffset2;
 746     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 747     int              vdwioffset3;
 748     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 749     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 750     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 751     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 752     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 753     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 754     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 755     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 756     real             *charge;
 757     int              nvdwtype;
 758     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 759     int              *vdwtype;
 760     real             *vdwparam;
 761     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 762     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 763     __m128i          vfitab;
 764     __m128i          ifour       = _mm_set1_epi32(4);
 765     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 766     real             *vftab;
 767     __m128           dummy_mask,cutoff_mask;
 768     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 769     __m128           one     = _mm_set1_ps(1.0);
 770     __m128           two     = _mm_set1_ps(2.0);
 771     x                = xx[0];
 772     f                = ff[0];
 773
 774     nri              = nlist->nri;
 775     iinr             = nlist->iinr;
 776     jindex           = nlist->jindex;
 777     jjnr             = nlist->jjnr;
 778     shiftidx         = nlist->shift;
 779     gid              = nlist->gid;
 780     shiftvec         = fr->shift_vec[0];
 781     fshift           = fr->fshift[0];
 782     facel            = _mm_set1_ps(fr->epsfac);
 783     charge           = mdatoms->chargeA;
 784     nvdwtype         = fr->ntype;
 785     vdwparam         = fr->nbfp;
 786     vdwtype          = mdatoms->typeA;
 787
 788     vftab            = kernel_data->table_elec->data;
 789     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 790
 791     /* Setup water-specific parameters */
 792     inr              = nlist->iinr[0];
 793     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 794     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 795     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 796     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 797
 798     /* Avoid stupid compiler warnings */
 799     jnrA = jnrB = jnrC = jnrD = 0;
 800     j_coord_offsetA = 0;
 801     j_coord_offsetB = 0;
 802     j_coord_offsetC = 0;
 803     j_coord_offsetD = 0;
 804
 805     outeriter        = 0;
 806     inneriter        = 0;
 807
 808     for(iidx=0;iidx<4*DIM;iidx++)
 809     {
 810         scratch[iidx] = 0.0;
 811     }
 812
 813     /* Start outer loop over neighborlists */
 814     for(iidx=0; iidx<nri; iidx++)
 815     {
 816         /* Load shift vector for this list */
 817         i_shift_offset   = DIM*shiftidx[iidx];
 818
 819         /* Load limits for loop over neighbors */
 820         j_index_start    = jindex[iidx];
 821         j_index_end      = jindex[iidx+1];
 822
 823         /* Get outer coordinate index */
 824         inr              = iinr[iidx];
 825         i_coord_offset   = DIM*inr;
 826
 827         /* Load i particle coords and add shift vector */
 828         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 829                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 830
 831         fix0             = _mm_setzero_ps();
 832         fiy0             = _mm_setzero_ps();
 833         fiz0             = _mm_setzero_ps();
 834         fix1             = _mm_setzero_ps();
 835         fiy1             = _mm_setzero_ps();
 836         fiz1             = _mm_setzero_ps();
 837         fix2             = _mm_setzero_ps();
 838         fiy2             = _mm_setzero_ps();
 839         fiz2             = _mm_setzero_ps();
 840         fix3             = _mm_setzero_ps();
 841         fiy3             = _mm_setzero_ps();
 842         fiz3             = _mm_setzero_ps();
 843
 844         /* Start inner kernel loop */
 845         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 846         {
 847
 848             /* Get j neighbor index, and coordinate index */
 849             jnrA             = jjnr[jidx];
 850             jnrB             = jjnr[jidx+1];
 851             jnrC             = jjnr[jidx+2];
 852             jnrD             = jjnr[jidx+3];
 853             j_coord_offsetA  = DIM*jnrA;
 854             j_coord_offsetB  = DIM*jnrB;
 855             j_coord_offsetC  = DIM*jnrC;
 856             j_coord_offsetD  = DIM*jnrD;
 857
 858             /* load j atom coordinates */
 859             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 860                                               x+j_coord_offsetC,x+j_coord_offsetD,
 861                                               &jx0,&jy0,&jz0);
 862
 863             /* Calculate displacement vector */
 864             dx00             = _mm_sub_ps(ix0,jx0);
 865             dy00             = _mm_sub_ps(iy0,jy0);
 866             dz00             = _mm_sub_ps(iz0,jz0);
 867             dx10             = _mm_sub_ps(ix1,jx0);
 868             dy10             = _mm_sub_ps(iy1,jy0);
 869             dz10             = _mm_sub_ps(iz1,jz0);
 870             dx20             = _mm_sub_ps(ix2,jx0);
 871             dy20             = _mm_sub_ps(iy2,jy0);
 872             dz20             = _mm_sub_ps(iz2,jz0);
 873             dx30             = _mm_sub_ps(ix3,jx0);
 874             dy30             = _mm_sub_ps(iy3,jy0);
 875             dz30             = _mm_sub_ps(iz3,jz0);
 876
 877             /* Calculate squared distance and things based on it */
 878             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 879             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 880             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 881             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 882
 883             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 884             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 885             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 886
 887             rinvsq00         = gmx_mm_inv_ps(rsq00);
 888
 889             /* Load parameters for j particles */
 890             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 891                                                               charge+jnrC+0,charge+jnrD+0);
 892             vdwjidx0A        = 2*vdwtype[jnrA+0];
 893             vdwjidx0B        = 2*vdwtype[jnrB+0];
 894             vdwjidx0C        = 2*vdwtype[jnrC+0];
 895             vdwjidx0D        = 2*vdwtype[jnrD+0];
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 903                                          vdwparam+vdwioffset0+vdwjidx0B,
 904                                          vdwparam+vdwioffset0+vdwjidx0C,
 905                                          vdwparam+vdwioffset0+vdwjidx0D,
 906                                          &c6_00,&c12_00);
 907
 908             /* LENNARD-JONES DISPERSION/REPULSION */
 909
 910             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 911             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 912
 913             fscal            = fvdw;
 914
 915             /* Calculate temporary vectorial force */
 916             tx               = _mm_mul_ps(fscal,dx00);
 917             ty               = _mm_mul_ps(fscal,dy00);
 918             tz               = _mm_mul_ps(fscal,dz00);
 919
 920             /* Update vectorial force */
 921             fix0             = _mm_add_ps(fix0,tx);
 922             fiy0             = _mm_add_ps(fiy0,ty);
 923             fiz0             = _mm_add_ps(fiz0,tz);
 924
 925             fjptrA             = f+j_coord_offsetA;
 926             fjptrB             = f+j_coord_offsetB;
 927             fjptrC             = f+j_coord_offsetC;
 928             fjptrD             = f+j_coord_offsetD;
 929             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 930
 931             /**************************
 932              * CALCULATE INTERACTIONS *
 933              **************************/
 934
 935             r10              = _mm_mul_ps(rsq10,rinv10);
 936
 937             /* Compute parameters for interactions between i and j atoms */
 938             qq10             = _mm_mul_ps(iq1,jq0);
 939
 940             /* Calculate table index by multiplying r with table scale and truncate to integer */
 941             rt               = _mm_mul_ps(r10,vftabscale);
 942             vfitab           = _mm_cvttps_epi32(rt);
 943             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 944             vfitab           = _mm_slli_epi32(vfitab,2);
 945
 946             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 947             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 948             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 949             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 950             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 951             _MM_TRANSPOSE4_PS(Y,F,G,H);
 952             Heps             = _mm_mul_ps(vfeps,H);
 953             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 954             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 955             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 956
 957             fscal            = felec;
 958
 959             /* Calculate temporary vectorial force */
 960             tx               = _mm_mul_ps(fscal,dx10);
 961             ty               = _mm_mul_ps(fscal,dy10);
 962             tz               = _mm_mul_ps(fscal,dz10);
 963
 964             /* Update vectorial force */
 965             fix1             = _mm_add_ps(fix1,tx);
 966             fiy1             = _mm_add_ps(fiy1,ty);
 967             fiz1             = _mm_add_ps(fiz1,tz);
 968
 969             fjptrA             = f+j_coord_offsetA;
 970             fjptrB             = f+j_coord_offsetB;
 971             fjptrC             = f+j_coord_offsetC;
 972             fjptrD             = f+j_coord_offsetD;
 973             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 974
 975             /**************************
 976              * CALCULATE INTERACTIONS *
 977              **************************/
 978
 979             r20              = _mm_mul_ps(rsq20,rinv20);
 980
 981             /* Compute parameters for interactions between i and j atoms */
 982             qq20             = _mm_mul_ps(iq2,jq0);
 983
 984             /* Calculate table index by multiplying r with table scale and truncate to integer */
 985             rt               = _mm_mul_ps(r20,vftabscale);
 986             vfitab           = _mm_cvttps_epi32(rt);
 987             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 988             vfitab           = _mm_slli_epi32(vfitab,2);
 989
 990             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 991             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 992             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 993             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 994             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 995             _MM_TRANSPOSE4_PS(Y,F,G,H);
 996             Heps             = _mm_mul_ps(vfeps,H);
 997             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 998             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 999             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1000
1001             fscal            = felec;
1002
1003             /* Calculate temporary vectorial force */
1004             tx               = _mm_mul_ps(fscal,dx20);
1005             ty               = _mm_mul_ps(fscal,dy20);
1006             tz               = _mm_mul_ps(fscal,dz20);
1007
1008             /* Update vectorial force */
1009             fix2             = _mm_add_ps(fix2,tx);
1010             fiy2             = _mm_add_ps(fiy2,ty);
1011             fiz2             = _mm_add_ps(fiz2,tz);
1012
1013             fjptrA             = f+j_coord_offsetA;
1014             fjptrB             = f+j_coord_offsetB;
1015             fjptrC             = f+j_coord_offsetC;
1016             fjptrD             = f+j_coord_offsetD;
1017             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             r30              = _mm_mul_ps(rsq30,rinv30);
1024
1025             /* Compute parameters for interactions between i and j atoms */
1026             qq30             = _mm_mul_ps(iq3,jq0);
1027
1028             /* Calculate table index by multiplying r with table scale and truncate to integer */
1029             rt               = _mm_mul_ps(r30,vftabscale);
1030             vfitab           = _mm_cvttps_epi32(rt);
1031             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1032             vfitab           = _mm_slli_epi32(vfitab,2);
1033
1034             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1035             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1036             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1037             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1038             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1039             _MM_TRANSPOSE4_PS(Y,F,G,H);
1040             Heps             = _mm_mul_ps(vfeps,H);
1041             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1042             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1043             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1044
1045             fscal            = felec;
1046
1047             /* Calculate temporary vectorial force */
1048             tx               = _mm_mul_ps(fscal,dx30);
1049             ty               = _mm_mul_ps(fscal,dy30);
1050             tz               = _mm_mul_ps(fscal,dz30);
1051
1052             /* Update vectorial force */
1053             fix3             = _mm_add_ps(fix3,tx);
1054             fiy3             = _mm_add_ps(fiy3,ty);
1055             fiz3             = _mm_add_ps(fiz3,tz);
1056
1057             fjptrA             = f+j_coord_offsetA;
1058             fjptrB             = f+j_coord_offsetB;
1059             fjptrC             = f+j_coord_offsetC;
1060             fjptrD             = f+j_coord_offsetD;
1061             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1062
1063             /* Inner loop uses 144 flops */
1064         }
1065
1066         if(jidx<j_index_end)
1067         {
1068
1069             /* Get j neighbor index, and coordinate index */
1070             jnrlistA         = jjnr[jidx];
1071             jnrlistB         = jjnr[jidx+1];
1072             jnrlistC         = jjnr[jidx+2];
1073             jnrlistD         = jjnr[jidx+3];
1074             /* Sign of each element will be negative for non-real atoms.
1075              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1076              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1077              */
1078             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1079             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1080             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1081             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1082             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1083             j_coord_offsetA  = DIM*jnrA;
1084             j_coord_offsetB  = DIM*jnrB;
1085             j_coord_offsetC  = DIM*jnrC;
1086             j_coord_offsetD  = DIM*jnrD;
1087
1088             /* load j atom coordinates */
1089             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1090                                               x+j_coord_offsetC,x+j_coord_offsetD,
1091                                               &jx0,&jy0,&jz0);
1092
1093             /* Calculate displacement vector */
1094             dx00             = _mm_sub_ps(ix0,jx0);
1095             dy00             = _mm_sub_ps(iy0,jy0);
1096             dz00             = _mm_sub_ps(iz0,jz0);
1097             dx10             = _mm_sub_ps(ix1,jx0);
1098             dy10             = _mm_sub_ps(iy1,jy0);
1099             dz10             = _mm_sub_ps(iz1,jz0);
1100             dx20             = _mm_sub_ps(ix2,jx0);
1101             dy20             = _mm_sub_ps(iy2,jy0);
1102             dz20             = _mm_sub_ps(iz2,jz0);
1103             dx30             = _mm_sub_ps(ix3,jx0);
1104             dy30             = _mm_sub_ps(iy3,jy0);
1105             dz30             = _mm_sub_ps(iz3,jz0);
1106
1107             /* Calculate squared distance and things based on it */
1108             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1109             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1110             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1111             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1112
1113             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1114             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1115             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1116
1117             rinvsq00         = gmx_mm_inv_ps(rsq00);
1118
1119             /* Load parameters for j particles */
1120             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1121                                                               charge+jnrC+0,charge+jnrD+0);
1122             vdwjidx0A        = 2*vdwtype[jnrA+0];
1123             vdwjidx0B        = 2*vdwtype[jnrB+0];
1124             vdwjidx0C        = 2*vdwtype[jnrC+0];
1125             vdwjidx0D        = 2*vdwtype[jnrD+0];
1126
1127             /**************************
1128              * CALCULATE INTERACTIONS *
1129              **************************/
1130
1131             /* Compute parameters for interactions between i and j atoms */
1132             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1133                                          vdwparam+vdwioffset0+vdwjidx0B,
1134                                          vdwparam+vdwioffset0+vdwjidx0C,
1135                                          vdwparam+vdwioffset0+vdwjidx0D,
1136                                          &c6_00,&c12_00);
1137
1138             /* LENNARD-JONES DISPERSION/REPULSION */
1139
1140             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1141             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1142
1143             fscal            = fvdw;
1144
1145             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1146
1147             /* Calculate temporary vectorial force */
1148             tx               = _mm_mul_ps(fscal,dx00);
1149             ty               = _mm_mul_ps(fscal,dy00);
1150             tz               = _mm_mul_ps(fscal,dz00);
1151
1152             /* Update vectorial force */
1153             fix0             = _mm_add_ps(fix0,tx);
1154             fiy0             = _mm_add_ps(fiy0,ty);
1155             fiz0             = _mm_add_ps(fiz0,tz);
1156
1157             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1158             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1159             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1160             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1161             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1162
1163             /**************************
1164              * CALCULATE INTERACTIONS *
1165              **************************/
1166
1167             r10              = _mm_mul_ps(rsq10,rinv10);
1168             r10              = _mm_andnot_ps(dummy_mask,r10);
1169
1170             /* Compute parameters for interactions between i and j atoms */
1171             qq10             = _mm_mul_ps(iq1,jq0);
1172
1173             /* Calculate table index by multiplying r with table scale and truncate to integer */
1174             rt               = _mm_mul_ps(r10,vftabscale);
1175             vfitab           = _mm_cvttps_epi32(rt);
1176             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1177             vfitab           = _mm_slli_epi32(vfitab,2);
1178
1179             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1180             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1181             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1182             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1183             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1184             _MM_TRANSPOSE4_PS(Y,F,G,H);
1185             Heps             = _mm_mul_ps(vfeps,H);
1186             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1187             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1188             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1189
1190             fscal            = felec;
1191
1192             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1193
1194             /* Calculate temporary vectorial force */
1195             tx               = _mm_mul_ps(fscal,dx10);
1196             ty               = _mm_mul_ps(fscal,dy10);
1197             tz               = _mm_mul_ps(fscal,dz10);
1198
1199             /* Update vectorial force */
1200             fix1             = _mm_add_ps(fix1,tx);
1201             fiy1             = _mm_add_ps(fiy1,ty);
1202             fiz1             = _mm_add_ps(fiz1,tz);
1203
1204             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1205             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1206             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1207             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1208             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1209
1210             /**************************
1211              * CALCULATE INTERACTIONS *
1212              **************************/
1213
1214             r20              = _mm_mul_ps(rsq20,rinv20);
1215             r20              = _mm_andnot_ps(dummy_mask,r20);
1216
1217             /* Compute parameters for interactions between i and j atoms */
1218             qq20             = _mm_mul_ps(iq2,jq0);
1219
1220             /* Calculate table index by multiplying r with table scale and truncate to integer */
1221             rt               = _mm_mul_ps(r20,vftabscale);
1222             vfitab           = _mm_cvttps_epi32(rt);
1223             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1224             vfitab           = _mm_slli_epi32(vfitab,2);
1225
1226             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1227             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1228             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1229             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1230             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1231             _MM_TRANSPOSE4_PS(Y,F,G,H);
1232             Heps             = _mm_mul_ps(vfeps,H);
1233             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1234             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1235             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1236
1237             fscal            = felec;
1238
1239             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1240
1241             /* Calculate temporary vectorial force */
1242             tx               = _mm_mul_ps(fscal,dx20);
1243             ty               = _mm_mul_ps(fscal,dy20);
1244             tz               = _mm_mul_ps(fscal,dz20);
1245
1246             /* Update vectorial force */
1247             fix2             = _mm_add_ps(fix2,tx);
1248             fiy2             = _mm_add_ps(fiy2,ty);
1249             fiz2             = _mm_add_ps(fiz2,tz);
1250
1251             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1252             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1253             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1254             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1255             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1256
1257             /**************************
1258              * CALCULATE INTERACTIONS *
1259              **************************/
1260
1261             r30              = _mm_mul_ps(rsq30,rinv30);
1262             r30              = _mm_andnot_ps(dummy_mask,r30);
1263
1264             /* Compute parameters for interactions between i and j atoms */
1265             qq30             = _mm_mul_ps(iq3,jq0);
1266
1267             /* Calculate table index by multiplying r with table scale and truncate to integer */
1268             rt               = _mm_mul_ps(r30,vftabscale);
1269             vfitab           = _mm_cvttps_epi32(rt);
1270             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1271             vfitab           = _mm_slli_epi32(vfitab,2);
1272
1273             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1274             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1275             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1276             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1277             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1278             _MM_TRANSPOSE4_PS(Y,F,G,H);
1279             Heps             = _mm_mul_ps(vfeps,H);
1280             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1281             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1282             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1283
1284             fscal            = felec;
1285
1286             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1287
1288             /* Calculate temporary vectorial force */
1289             tx               = _mm_mul_ps(fscal,dx30);
1290             ty               = _mm_mul_ps(fscal,dy30);
1291             tz               = _mm_mul_ps(fscal,dz30);
1292
1293             /* Update vectorial force */
1294             fix3             = _mm_add_ps(fix3,tx);
1295             fiy3             = _mm_add_ps(fiy3,ty);
1296             fiz3             = _mm_add_ps(fiz3,tz);
1297
1298             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1299             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1300             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1301             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1302             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1303
1304             /* Inner loop uses 147 flops */
1305         }
1306
1307         /* End of innermost loop */
1308
1309         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1310                                               f+i_coord_offset,fshift+i_shift_offset);
1311
1312         /* Increment number of inner iterations */
1313         inneriter                  += j_index_end - j_index_start;
1314
1315         /* Outer loop uses 24 flops */
1316     }
1317
1318     /* Increment number of outer iterations */
1319     outeriter        += nri;
1320
1321     /* Update outer/inner flops */
1322
1323     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1324 }