src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse4_1_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse4_1_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              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     krf              = _mm_set1_ps(fr->ic->k_rf);
 113     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_ps(fr->ic->c_rf);
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_vdw->data;
 120     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 121
 122     /* Setup water-specific parameters */
 123     inr              = nlist->iinr[0];
 124     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 125     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 126     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 127     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 128
 129     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 130     rcutoff_scalar   = fr->rcoulomb;
 131     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 132     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = jnrC = jnrD = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138     j_coord_offsetC = 0;
 139     j_coord_offsetD = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 165                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 166
 167         fix0             = _mm_setzero_ps();
 168         fiy0             = _mm_setzero_ps();
 169         fiz0             = _mm_setzero_ps();
 170         fix1             = _mm_setzero_ps();
 171         fiy1             = _mm_setzero_ps();
 172         fiz1             = _mm_setzero_ps();
 173         fix2             = _mm_setzero_ps();
 174         fiy2             = _mm_setzero_ps();
 175         fiz2             = _mm_setzero_ps();
 176         fix3             = _mm_setzero_ps();
 177         fiy3             = _mm_setzero_ps();
 178         fiz3             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213             dx30             = _mm_sub_ps(ix3,jx0);
 214             dy30             = _mm_sub_ps(iy3,jy0);
 215             dz30             = _mm_sub_ps(iz3,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 219             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 220             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 221             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 222
 223             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 224             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 226             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 227
 228             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 230             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 234                                                               charge+jnrC+0,charge+jnrD+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237             vdwjidx0C        = 2*vdwtype[jnrC+0];
 238             vdwjidx0D        = 2*vdwtype[jnrD+0];
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             r00              = _mm_mul_ps(rsq00,rinv00);
 245
 246             /* Compute parameters for interactions between i and j atoms */
 247             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 248                                          vdwparam+vdwioffset0+vdwjidx0B,
 249                                          vdwparam+vdwioffset0+vdwjidx0C,
 250                                          vdwparam+vdwioffset0+vdwjidx0D,
 251                                          &c6_00,&c12_00);
 252
 253             /* Calculate table index by multiplying r with table scale and truncate to integer */
 254             rt               = _mm_mul_ps(r00,vftabscale);
 255             vfitab           = _mm_cvttps_epi32(rt);
 256             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 257             vfitab           = _mm_slli_epi32(vfitab,3);
 258
 259             /* CUBIC SPLINE TABLE DISPERSION */
 260             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 261             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 262             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 263             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 264             _MM_TRANSPOSE4_PS(Y,F,G,H);
 265             Heps             = _mm_mul_ps(vfeps,H);
 266             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 267             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 268             vvdw6            = _mm_mul_ps(c6_00,VV);
 269             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 270             fvdw6            = _mm_mul_ps(c6_00,FF);
 271
 272             /* CUBIC SPLINE TABLE REPULSION */
 273             vfitab           = _mm_add_epi32(vfitab,ifour);
 274             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 275             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 276             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 277             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 278             _MM_TRANSPOSE4_PS(Y,F,G,H);
 279             Heps             = _mm_mul_ps(vfeps,H);
 280             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 281             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 282             vvdw12           = _mm_mul_ps(c12_00,VV);
 283             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 284             fvdw12           = _mm_mul_ps(c12_00,FF);
 285             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 286             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 290
 291             fscal            = fvdw;
 292
 293             /* Calculate temporary vectorial force */
 294             tx               = _mm_mul_ps(fscal,dx00);
 295             ty               = _mm_mul_ps(fscal,dy00);
 296             tz               = _mm_mul_ps(fscal,dz00);
 297
 298             /* Update vectorial force */
 299             fix0             = _mm_add_ps(fix0,tx);
 300             fiy0             = _mm_add_ps(fiy0,ty);
 301             fiz0             = _mm_add_ps(fiz0,tz);
 302
 303             fjptrA             = f+j_coord_offsetA;
 304             fjptrB             = f+j_coord_offsetB;
 305             fjptrC             = f+j_coord_offsetC;
 306             fjptrD             = f+j_coord_offsetD;
 307             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 308
 309             /**************************
 310              * CALCULATE INTERACTIONS *
 311              **************************/
 312
 313             if (gmx_mm_any_lt(rsq10,rcutoff2))
 314             {
 315
 316             /* Compute parameters for interactions between i and j atoms */
 317             qq10             = _mm_mul_ps(iq1,jq0);
 318
 319             /* REACTION-FIELD ELECTROSTATICS */
 320             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 321             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 322
 323             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velec            = _mm_and_ps(velec,cutoff_mask);
 327             velecsum         = _mm_add_ps(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             fscal            = _mm_and_ps(fscal,cutoff_mask);
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_ps(fscal,dx10);
 335             ty               = _mm_mul_ps(fscal,dy10);
 336             tz               = _mm_mul_ps(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm_add_ps(fix1,tx);
 340             fiy1             = _mm_add_ps(fiy1,ty);
 341             fiz1             = _mm_add_ps(fiz1,tz);
 342
 343             fjptrA             = f+j_coord_offsetA;
 344             fjptrB             = f+j_coord_offsetB;
 345             fjptrC             = f+j_coord_offsetC;
 346             fjptrD             = f+j_coord_offsetD;
 347             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 348
 349             }
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             if (gmx_mm_any_lt(rsq20,rcutoff2))
 356             {
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq20             = _mm_mul_ps(iq2,jq0);
 360
 361             /* REACTION-FIELD ELECTROSTATICS */
 362             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 363             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 364
 365             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velec            = _mm_and_ps(velec,cutoff_mask);
 369             velecsum         = _mm_add_ps(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             fscal            = _mm_and_ps(fscal,cutoff_mask);
 374
 375             /* Calculate temporary vectorial force */
 376             tx               = _mm_mul_ps(fscal,dx20);
 377             ty               = _mm_mul_ps(fscal,dy20);
 378             tz               = _mm_mul_ps(fscal,dz20);
 379
 380             /* Update vectorial force */
 381             fix2             = _mm_add_ps(fix2,tx);
 382             fiy2             = _mm_add_ps(fiy2,ty);
 383             fiz2             = _mm_add_ps(fiz2,tz);
 384
 385             fjptrA             = f+j_coord_offsetA;
 386             fjptrB             = f+j_coord_offsetB;
 387             fjptrC             = f+j_coord_offsetC;
 388             fjptrD             = f+j_coord_offsetD;
 389             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 390
 391             }
 392
 393             /**************************
 394              * CALCULATE INTERACTIONS *
 395              **************************/
 396
 397             if (gmx_mm_any_lt(rsq30,rcutoff2))
 398             {
 399
 400             /* Compute parameters for interactions between i and j atoms */
 401             qq30             = _mm_mul_ps(iq3,jq0);
 402
 403             /* REACTION-FIELD ELECTROSTATICS */
 404             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 405             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 406
 407             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velec            = _mm_and_ps(velec,cutoff_mask);
 411             velecsum         = _mm_add_ps(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             fscal            = _mm_and_ps(fscal,cutoff_mask);
 416
 417             /* Calculate temporary vectorial force */
 418             tx               = _mm_mul_ps(fscal,dx30);
 419             ty               = _mm_mul_ps(fscal,dy30);
 420             tz               = _mm_mul_ps(fscal,dz30);
 421
 422             /* Update vectorial force */
 423             fix3             = _mm_add_ps(fix3,tx);
 424             fiy3             = _mm_add_ps(fiy3,ty);
 425             fiz3             = _mm_add_ps(fiz3,tz);
 426
 427             fjptrA             = f+j_coord_offsetA;
 428             fjptrB             = f+j_coord_offsetB;
 429             fjptrC             = f+j_coord_offsetC;
 430             fjptrD             = f+j_coord_offsetD;
 431             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 432
 433             }
 434
 435             /* Inner loop uses 164 flops */
 436         }
 437
 438         if(jidx<j_index_end)
 439         {
 440
 441             /* Get j neighbor index, and coordinate index */
 442             jnrlistA         = jjnr[jidx];
 443             jnrlistB         = jjnr[jidx+1];
 444             jnrlistC         = jjnr[jidx+2];
 445             jnrlistD         = jjnr[jidx+3];
 446             /* Sign of each element will be negative for non-real atoms.
 447              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 448              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 449              */
 450             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 451             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 452             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 453             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 454             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 455             j_coord_offsetA  = DIM*jnrA;
 456             j_coord_offsetB  = DIM*jnrB;
 457             j_coord_offsetC  = DIM*jnrC;
 458             j_coord_offsetD  = DIM*jnrD;
 459
 460             /* load j atom coordinates */
 461             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 462                                               x+j_coord_offsetC,x+j_coord_offsetD,
 463                                               &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm_sub_ps(ix0,jx0);
 467             dy00             = _mm_sub_ps(iy0,jy0);
 468             dz00             = _mm_sub_ps(iz0,jz0);
 469             dx10             = _mm_sub_ps(ix1,jx0);
 470             dy10             = _mm_sub_ps(iy1,jy0);
 471             dz10             = _mm_sub_ps(iz1,jz0);
 472             dx20             = _mm_sub_ps(ix2,jx0);
 473             dy20             = _mm_sub_ps(iy2,jy0);
 474             dz20             = _mm_sub_ps(iz2,jz0);
 475             dx30             = _mm_sub_ps(ix3,jx0);
 476             dy30             = _mm_sub_ps(iy3,jy0);
 477             dz30             = _mm_sub_ps(iz3,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 481             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 482             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 483             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 484
 485             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 486             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 487             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 488             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 489
 490             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 491             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 492             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 493
 494             /* Load parameters for j particles */
 495             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 496                                                               charge+jnrC+0,charge+jnrD+0);
 497             vdwjidx0A        = 2*vdwtype[jnrA+0];
 498             vdwjidx0B        = 2*vdwtype[jnrB+0];
 499             vdwjidx0C        = 2*vdwtype[jnrC+0];
 500             vdwjidx0D        = 2*vdwtype[jnrD+0];
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             r00              = _mm_mul_ps(rsq00,rinv00);
 507             r00              = _mm_andnot_ps(dummy_mask,r00);
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 511                                          vdwparam+vdwioffset0+vdwjidx0B,
 512                                          vdwparam+vdwioffset0+vdwjidx0C,
 513                                          vdwparam+vdwioffset0+vdwjidx0D,
 514                                          &c6_00,&c12_00);
 515
 516             /* Calculate table index by multiplying r with table scale and truncate to integer */
 517             rt               = _mm_mul_ps(r00,vftabscale);
 518             vfitab           = _mm_cvttps_epi32(rt);
 519             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 520             vfitab           = _mm_slli_epi32(vfitab,3);
 521
 522             /* CUBIC SPLINE TABLE DISPERSION */
 523             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 524             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 525             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 526             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 527             _MM_TRANSPOSE4_PS(Y,F,G,H);
 528             Heps             = _mm_mul_ps(vfeps,H);
 529             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 530             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 531             vvdw6            = _mm_mul_ps(c6_00,VV);
 532             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 533             fvdw6            = _mm_mul_ps(c6_00,FF);
 534
 535             /* CUBIC SPLINE TABLE REPULSION */
 536             vfitab           = _mm_add_epi32(vfitab,ifour);
 537             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 538             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 539             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 540             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 541             _MM_TRANSPOSE4_PS(Y,F,G,H);
 542             Heps             = _mm_mul_ps(vfeps,H);
 543             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 544             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 545             vvdw12           = _mm_mul_ps(c12_00,VV);
 546             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 547             fvdw12           = _mm_mul_ps(c12_00,FF);
 548             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 549             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 553             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 554
 555             fscal            = fvdw;
 556
 557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm_mul_ps(fscal,dx00);
 561             ty               = _mm_mul_ps(fscal,dy00);
 562             tz               = _mm_mul_ps(fscal,dz00);
 563
 564             /* Update vectorial force */
 565             fix0             = _mm_add_ps(fix0,tx);
 566             fiy0             = _mm_add_ps(fiy0,ty);
 567             fiz0             = _mm_add_ps(fiz0,tz);
 568
 569             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 570             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 571             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 572             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 573             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             if (gmx_mm_any_lt(rsq10,rcutoff2))
 580             {
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq10             = _mm_mul_ps(iq1,jq0);
 584
 585             /* REACTION-FIELD ELECTROSTATICS */
 586             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 587             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 588
 589             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velec            = _mm_and_ps(velec,cutoff_mask);
 593             velec            = _mm_andnot_ps(dummy_mask,velec);
 594             velecsum         = _mm_add_ps(velecsum,velec);
 595
 596             fscal            = felec;
 597
 598             fscal            = _mm_and_ps(fscal,cutoff_mask);
 599
 600             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 601
 602             /* Calculate temporary vectorial force */
 603             tx               = _mm_mul_ps(fscal,dx10);
 604             ty               = _mm_mul_ps(fscal,dy10);
 605             tz               = _mm_mul_ps(fscal,dz10);
 606
 607             /* Update vectorial force */
 608             fix1             = _mm_add_ps(fix1,tx);
 609             fiy1             = _mm_add_ps(fiy1,ty);
 610             fiz1             = _mm_add_ps(fiz1,tz);
 611
 612             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 613             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 614             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 615             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 616             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 617
 618             }
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             if (gmx_mm_any_lt(rsq20,rcutoff2))
 625             {
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq20             = _mm_mul_ps(iq2,jq0);
 629
 630             /* REACTION-FIELD ELECTROSTATICS */
 631             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 632             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 633
 634             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 635
 636             /* Update potential sum for this i atom from the interaction with this j atom. */
 637             velec            = _mm_and_ps(velec,cutoff_mask);
 638             velec            = _mm_andnot_ps(dummy_mask,velec);
 639             velecsum         = _mm_add_ps(velecsum,velec);
 640
 641             fscal            = felec;
 642
 643             fscal            = _mm_and_ps(fscal,cutoff_mask);
 644
 645             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm_mul_ps(fscal,dx20);
 649             ty               = _mm_mul_ps(fscal,dy20);
 650             tz               = _mm_mul_ps(fscal,dz20);
 651
 652             /* Update vectorial force */
 653             fix2             = _mm_add_ps(fix2,tx);
 654             fiy2             = _mm_add_ps(fiy2,ty);
 655             fiz2             = _mm_add_ps(fiz2,tz);
 656
 657             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 658             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 659             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 660             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 661             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 662
 663             }
 664
 665             /**************************
 666              * CALCULATE INTERACTIONS *
 667              **************************/
 668
 669             if (gmx_mm_any_lt(rsq30,rcutoff2))
 670             {
 671
 672             /* Compute parameters for interactions between i and j atoms */
 673             qq30             = _mm_mul_ps(iq3,jq0);
 674
 675             /* REACTION-FIELD ELECTROSTATICS */
 676             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 677             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 678
 679             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 680
 681             /* Update potential sum for this i atom from the interaction with this j atom. */
 682             velec            = _mm_and_ps(velec,cutoff_mask);
 683             velec            = _mm_andnot_ps(dummy_mask,velec);
 684             velecsum         = _mm_add_ps(velecsum,velec);
 685
 686             fscal            = felec;
 687
 688             fscal            = _mm_and_ps(fscal,cutoff_mask);
 689
 690             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 691
 692             /* Calculate temporary vectorial force */
 693             tx               = _mm_mul_ps(fscal,dx30);
 694             ty               = _mm_mul_ps(fscal,dy30);
 695             tz               = _mm_mul_ps(fscal,dz30);
 696
 697             /* Update vectorial force */
 698             fix3             = _mm_add_ps(fix3,tx);
 699             fiy3             = _mm_add_ps(fiy3,ty);
 700             fiz3             = _mm_add_ps(fiz3,tz);
 701
 702             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 703             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 704             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 705             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 706             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 707
 708             }
 709
 710             /* Inner loop uses 165 flops */
 711         }
 712
 713         /* End of innermost loop */
 714
 715         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 716                                               f+i_coord_offset,fshift+i_shift_offset);
 717
 718         ggid                        = gid[iidx];
 719         /* Update potential energies */
 720         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 721         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 722
 723         /* Increment number of inner iterations */
 724         inneriter                  += j_index_end - j_index_start;
 725
 726         /* Outer loop uses 26 flops */
 727     }
 728
 729     /* Increment number of outer iterations */
 730     outeriter        += nri;
 731
 732     /* Update outer/inner flops */
 733
 734     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*165);
 735 }
 736 /*
 737  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_single
 738  * Electrostatics interaction: ReactionField
 739  * VdW interaction:            CubicSplineTable
 740  * Geometry:                   Water4-Particle
 741  * Calculate force/pot:        Force
 742  */
 743 void
 744 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_single
 745                     (t_nblist * gmx_restrict                nlist,
 746                      rvec * gmx_restrict                    xx,
 747                      rvec * gmx_restrict                    ff,
 748                      t_forcerec * gmx_restrict              fr,
 749                      t_mdatoms * gmx_restrict               mdatoms,
 750                      nb_kernel_data_t * gmx_restrict        kernel_data,
 751                      t_nrnb * gmx_restrict                  nrnb)
 752 {
 753     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 754      * just 0 for non-waters.
 755      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 756      * jnr indices corresponding to data put in the four positions in the SIMD register.
 757      */
 758     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 759     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 760     int              jnrA,jnrB,jnrC,jnrD;
 761     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 762     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 763     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 764     real             rcutoff_scalar;
 765     real             *shiftvec,*fshift,*x,*f;
 766     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 767     real             scratch[4*DIM];
 768     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 769     int              vdwioffset0;
 770     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 771     int              vdwioffset1;
 772     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 773     int              vdwioffset2;
 774     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 775     int              vdwioffset3;
 776     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 777     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 778     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 779     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 780     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 781     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 782     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 783     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 784     real             *charge;
 785     int              nvdwtype;
 786     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 787     int              *vdwtype;
 788     real             *vdwparam;
 789     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 790     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 791     __m128i          vfitab;
 792     __m128i          ifour       = _mm_set1_epi32(4);
 793     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 794     real             *vftab;
 795     __m128           dummy_mask,cutoff_mask;
 796     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 797     __m128           one     = _mm_set1_ps(1.0);
 798     __m128           two     = _mm_set1_ps(2.0);
 799     x                = xx[0];
 800     f                = ff[0];
 801
 802     nri              = nlist->nri;
 803     iinr             = nlist->iinr;
 804     jindex           = nlist->jindex;
 805     jjnr             = nlist->jjnr;
 806     shiftidx         = nlist->shift;
 807     gid              = nlist->gid;
 808     shiftvec         = fr->shift_vec[0];
 809     fshift           = fr->fshift[0];
 810     facel            = _mm_set1_ps(fr->epsfac);
 811     charge           = mdatoms->chargeA;
 812     krf              = _mm_set1_ps(fr->ic->k_rf);
 813     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 814     crf              = _mm_set1_ps(fr->ic->c_rf);
 815     nvdwtype         = fr->ntype;
 816     vdwparam         = fr->nbfp;
 817     vdwtype          = mdatoms->typeA;
 818
 819     vftab            = kernel_data->table_vdw->data;
 820     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 821
 822     /* Setup water-specific parameters */
 823     inr              = nlist->iinr[0];
 824     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 825     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 826     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 827     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 828
 829     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 830     rcutoff_scalar   = fr->rcoulomb;
 831     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 832     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 833
 834     /* Avoid stupid compiler warnings */
 835     jnrA = jnrB = jnrC = jnrD = 0;
 836     j_coord_offsetA = 0;
 837     j_coord_offsetB = 0;
 838     j_coord_offsetC = 0;
 839     j_coord_offsetD = 0;
 840
 841     outeriter        = 0;
 842     inneriter        = 0;
 843
 844     for(iidx=0;iidx<4*DIM;iidx++)
 845     {
 846         scratch[iidx] = 0.0;
 847     }
 848
 849     /* Start outer loop over neighborlists */
 850     for(iidx=0; iidx<nri; iidx++)
 851     {
 852         /* Load shift vector for this list */
 853         i_shift_offset   = DIM*shiftidx[iidx];
 854
 855         /* Load limits for loop over neighbors */
 856         j_index_start    = jindex[iidx];
 857         j_index_end      = jindex[iidx+1];
 858
 859         /* Get outer coordinate index */
 860         inr              = iinr[iidx];
 861         i_coord_offset   = DIM*inr;
 862
 863         /* Load i particle coords and add shift vector */
 864         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 865                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 866
 867         fix0             = _mm_setzero_ps();
 868         fiy0             = _mm_setzero_ps();
 869         fiz0             = _mm_setzero_ps();
 870         fix1             = _mm_setzero_ps();
 871         fiy1             = _mm_setzero_ps();
 872         fiz1             = _mm_setzero_ps();
 873         fix2             = _mm_setzero_ps();
 874         fiy2             = _mm_setzero_ps();
 875         fiz2             = _mm_setzero_ps();
 876         fix3             = _mm_setzero_ps();
 877         fiy3             = _mm_setzero_ps();
 878         fiz3             = _mm_setzero_ps();
 879
 880         /* Start inner kernel loop */
 881         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 882         {
 883
 884             /* Get j neighbor index, and coordinate index */
 885             jnrA             = jjnr[jidx];
 886             jnrB             = jjnr[jidx+1];
 887             jnrC             = jjnr[jidx+2];
 888             jnrD             = jjnr[jidx+3];
 889             j_coord_offsetA  = DIM*jnrA;
 890             j_coord_offsetB  = DIM*jnrB;
 891             j_coord_offsetC  = DIM*jnrC;
 892             j_coord_offsetD  = DIM*jnrD;
 893
 894             /* load j atom coordinates */
 895             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 896                                               x+j_coord_offsetC,x+j_coord_offsetD,
 897                                               &jx0,&jy0,&jz0);
 898
 899             /* Calculate displacement vector */
 900             dx00             = _mm_sub_ps(ix0,jx0);
 901             dy00             = _mm_sub_ps(iy0,jy0);
 902             dz00             = _mm_sub_ps(iz0,jz0);
 903             dx10             = _mm_sub_ps(ix1,jx0);
 904             dy10             = _mm_sub_ps(iy1,jy0);
 905             dz10             = _mm_sub_ps(iz1,jz0);
 906             dx20             = _mm_sub_ps(ix2,jx0);
 907             dy20             = _mm_sub_ps(iy2,jy0);
 908             dz20             = _mm_sub_ps(iz2,jz0);
 909             dx30             = _mm_sub_ps(ix3,jx0);
 910             dy30             = _mm_sub_ps(iy3,jy0);
 911             dz30             = _mm_sub_ps(iz3,jz0);
 912
 913             /* Calculate squared distance and things based on it */
 914             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 915             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 916             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 917             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 918
 919             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 920             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 921             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 922             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 923
 924             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 925             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 926             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 927
 928             /* Load parameters for j particles */
 929             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 930                                                               charge+jnrC+0,charge+jnrD+0);
 931             vdwjidx0A        = 2*vdwtype[jnrA+0];
 932             vdwjidx0B        = 2*vdwtype[jnrB+0];
 933             vdwjidx0C        = 2*vdwtype[jnrC+0];
 934             vdwjidx0D        = 2*vdwtype[jnrD+0];
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             r00              = _mm_mul_ps(rsq00,rinv00);
 941
 942             /* Compute parameters for interactions between i and j atoms */
 943             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 944                                          vdwparam+vdwioffset0+vdwjidx0B,
 945                                          vdwparam+vdwioffset0+vdwjidx0C,
 946                                          vdwparam+vdwioffset0+vdwjidx0D,
 947                                          &c6_00,&c12_00);
 948
 949             /* Calculate table index by multiplying r with table scale and truncate to integer */
 950             rt               = _mm_mul_ps(r00,vftabscale);
 951             vfitab           = _mm_cvttps_epi32(rt);
 952             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 953             vfitab           = _mm_slli_epi32(vfitab,3);
 954
 955             /* CUBIC SPLINE TABLE DISPERSION */
 956             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 957             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 958             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 959             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 960             _MM_TRANSPOSE4_PS(Y,F,G,H);
 961             Heps             = _mm_mul_ps(vfeps,H);
 962             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 963             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 964             fvdw6            = _mm_mul_ps(c6_00,FF);
 965
 966             /* CUBIC SPLINE TABLE REPULSION */
 967             vfitab           = _mm_add_epi32(vfitab,ifour);
 968             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 969             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 970             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 971             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 972             _MM_TRANSPOSE4_PS(Y,F,G,H);
 973             Heps             = _mm_mul_ps(vfeps,H);
 974             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 975             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 976             fvdw12           = _mm_mul_ps(c12_00,FF);
 977             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 978
 979             fscal            = fvdw;
 980
 981             /* Calculate temporary vectorial force */
 982             tx               = _mm_mul_ps(fscal,dx00);
 983             ty               = _mm_mul_ps(fscal,dy00);
 984             tz               = _mm_mul_ps(fscal,dz00);
 985
 986             /* Update vectorial force */
 987             fix0             = _mm_add_ps(fix0,tx);
 988             fiy0             = _mm_add_ps(fiy0,ty);
 989             fiz0             = _mm_add_ps(fiz0,tz);
 990
 991             fjptrA             = f+j_coord_offsetA;
 992             fjptrB             = f+j_coord_offsetB;
 993             fjptrC             = f+j_coord_offsetC;
 994             fjptrD             = f+j_coord_offsetD;
 995             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             if (gmx_mm_any_lt(rsq10,rcutoff2))
1002             {
1003
1004             /* Compute parameters for interactions between i and j atoms */
1005             qq10             = _mm_mul_ps(iq1,jq0);
1006
1007             /* REACTION-FIELD ELECTROSTATICS */
1008             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1009
1010             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1011
1012             fscal            = felec;
1013
1014             fscal            = _mm_and_ps(fscal,cutoff_mask);
1015
1016             /* Calculate temporary vectorial force */
1017             tx               = _mm_mul_ps(fscal,dx10);
1018             ty               = _mm_mul_ps(fscal,dy10);
1019             tz               = _mm_mul_ps(fscal,dz10);
1020
1021             /* Update vectorial force */
1022             fix1             = _mm_add_ps(fix1,tx);
1023             fiy1             = _mm_add_ps(fiy1,ty);
1024             fiz1             = _mm_add_ps(fiz1,tz);
1025
1026             fjptrA             = f+j_coord_offsetA;
1027             fjptrB             = f+j_coord_offsetB;
1028             fjptrC             = f+j_coord_offsetC;
1029             fjptrD             = f+j_coord_offsetD;
1030             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1031
1032             }
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             if (gmx_mm_any_lt(rsq20,rcutoff2))
1039             {
1040
1041             /* Compute parameters for interactions between i and j atoms */
1042             qq20             = _mm_mul_ps(iq2,jq0);
1043
1044             /* REACTION-FIELD ELECTROSTATICS */
1045             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1046
1047             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1048
1049             fscal            = felec;
1050
1051             fscal            = _mm_and_ps(fscal,cutoff_mask);
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm_mul_ps(fscal,dx20);
1055             ty               = _mm_mul_ps(fscal,dy20);
1056             tz               = _mm_mul_ps(fscal,dz20);
1057
1058             /* Update vectorial force */
1059             fix2             = _mm_add_ps(fix2,tx);
1060             fiy2             = _mm_add_ps(fiy2,ty);
1061             fiz2             = _mm_add_ps(fiz2,tz);
1062
1063             fjptrA             = f+j_coord_offsetA;
1064             fjptrB             = f+j_coord_offsetB;
1065             fjptrC             = f+j_coord_offsetC;
1066             fjptrD             = f+j_coord_offsetD;
1067             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1068
1069             }
1070
1071             /**************************
1072              * CALCULATE INTERACTIONS *
1073              **************************/
1074
1075             if (gmx_mm_any_lt(rsq30,rcutoff2))
1076             {
1077
1078             /* Compute parameters for interactions between i and j atoms */
1079             qq30             = _mm_mul_ps(iq3,jq0);
1080
1081             /* REACTION-FIELD ELECTROSTATICS */
1082             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1083
1084             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1085
1086             fscal            = felec;
1087
1088             fscal            = _mm_and_ps(fscal,cutoff_mask);
1089
1090             /* Calculate temporary vectorial force */
1091             tx               = _mm_mul_ps(fscal,dx30);
1092             ty               = _mm_mul_ps(fscal,dy30);
1093             tz               = _mm_mul_ps(fscal,dz30);
1094
1095             /* Update vectorial force */
1096             fix3             = _mm_add_ps(fix3,tx);
1097             fiy3             = _mm_add_ps(fiy3,ty);
1098             fiz3             = _mm_add_ps(fiz3,tz);
1099
1100             fjptrA             = f+j_coord_offsetA;
1101             fjptrB             = f+j_coord_offsetB;
1102             fjptrC             = f+j_coord_offsetC;
1103             fjptrD             = f+j_coord_offsetD;
1104             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1105
1106             }
1107
1108             /* Inner loop uses 138 flops */
1109         }
1110
1111         if(jidx<j_index_end)
1112         {
1113
1114             /* Get j neighbor index, and coordinate index */
1115             jnrlistA         = jjnr[jidx];
1116             jnrlistB         = jjnr[jidx+1];
1117             jnrlistC         = jjnr[jidx+2];
1118             jnrlistD         = jjnr[jidx+3];
1119             /* Sign of each element will be negative for non-real atoms.
1120              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1121              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1122              */
1123             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1124             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1125             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1126             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1127             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1128             j_coord_offsetA  = DIM*jnrA;
1129             j_coord_offsetB  = DIM*jnrB;
1130             j_coord_offsetC  = DIM*jnrC;
1131             j_coord_offsetD  = DIM*jnrD;
1132
1133             /* load j atom coordinates */
1134             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1135                                               x+j_coord_offsetC,x+j_coord_offsetD,
1136                                               &jx0,&jy0,&jz0);
1137
1138             /* Calculate displacement vector */
1139             dx00             = _mm_sub_ps(ix0,jx0);
1140             dy00             = _mm_sub_ps(iy0,jy0);
1141             dz00             = _mm_sub_ps(iz0,jz0);
1142             dx10             = _mm_sub_ps(ix1,jx0);
1143             dy10             = _mm_sub_ps(iy1,jy0);
1144             dz10             = _mm_sub_ps(iz1,jz0);
1145             dx20             = _mm_sub_ps(ix2,jx0);
1146             dy20             = _mm_sub_ps(iy2,jy0);
1147             dz20             = _mm_sub_ps(iz2,jz0);
1148             dx30             = _mm_sub_ps(ix3,jx0);
1149             dy30             = _mm_sub_ps(iy3,jy0);
1150             dz30             = _mm_sub_ps(iz3,jz0);
1151
1152             /* Calculate squared distance and things based on it */
1153             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1154             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1155             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1156             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1157
1158             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1159             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1160             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1161             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1162
1163             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1164             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1165             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1166
1167             /* Load parameters for j particles */
1168             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1169                                                               charge+jnrC+0,charge+jnrD+0);
1170             vdwjidx0A        = 2*vdwtype[jnrA+0];
1171             vdwjidx0B        = 2*vdwtype[jnrB+0];
1172             vdwjidx0C        = 2*vdwtype[jnrC+0];
1173             vdwjidx0D        = 2*vdwtype[jnrD+0];
1174
1175             /**************************
1176              * CALCULATE INTERACTIONS *
1177              **************************/
1178
1179             r00              = _mm_mul_ps(rsq00,rinv00);
1180             r00              = _mm_andnot_ps(dummy_mask,r00);
1181
1182             /* Compute parameters for interactions between i and j atoms */
1183             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1184                                          vdwparam+vdwioffset0+vdwjidx0B,
1185                                          vdwparam+vdwioffset0+vdwjidx0C,
1186                                          vdwparam+vdwioffset0+vdwjidx0D,
1187                                          &c6_00,&c12_00);
1188
1189             /* Calculate table index by multiplying r with table scale and truncate to integer */
1190             rt               = _mm_mul_ps(r00,vftabscale);
1191             vfitab           = _mm_cvttps_epi32(rt);
1192             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1193             vfitab           = _mm_slli_epi32(vfitab,3);
1194
1195             /* CUBIC SPLINE TABLE DISPERSION */
1196             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1197             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1198             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1199             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1200             _MM_TRANSPOSE4_PS(Y,F,G,H);
1201             Heps             = _mm_mul_ps(vfeps,H);
1202             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1203             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1204             fvdw6            = _mm_mul_ps(c6_00,FF);
1205
1206             /* CUBIC SPLINE TABLE REPULSION */
1207             vfitab           = _mm_add_epi32(vfitab,ifour);
1208             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1209             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1210             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1211             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1212             _MM_TRANSPOSE4_PS(Y,F,G,H);
1213             Heps             = _mm_mul_ps(vfeps,H);
1214             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1215             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1216             fvdw12           = _mm_mul_ps(c12_00,FF);
1217             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1218
1219             fscal            = fvdw;
1220
1221             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1222
1223             /* Calculate temporary vectorial force */
1224             tx               = _mm_mul_ps(fscal,dx00);
1225             ty               = _mm_mul_ps(fscal,dy00);
1226             tz               = _mm_mul_ps(fscal,dz00);
1227
1228             /* Update vectorial force */
1229             fix0             = _mm_add_ps(fix0,tx);
1230             fiy0             = _mm_add_ps(fiy0,ty);
1231             fiz0             = _mm_add_ps(fiz0,tz);
1232
1233             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1234             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1235             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1236             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1237             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             if (gmx_mm_any_lt(rsq10,rcutoff2))
1244             {
1245
1246             /* Compute parameters for interactions between i and j atoms */
1247             qq10             = _mm_mul_ps(iq1,jq0);
1248
1249             /* REACTION-FIELD ELECTROSTATICS */
1250             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1251
1252             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1253
1254             fscal            = felec;
1255
1256             fscal            = _mm_and_ps(fscal,cutoff_mask);
1257
1258             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1259
1260             /* Calculate temporary vectorial force */
1261             tx               = _mm_mul_ps(fscal,dx10);
1262             ty               = _mm_mul_ps(fscal,dy10);
1263             tz               = _mm_mul_ps(fscal,dz10);
1264
1265             /* Update vectorial force */
1266             fix1             = _mm_add_ps(fix1,tx);
1267             fiy1             = _mm_add_ps(fiy1,ty);
1268             fiz1             = _mm_add_ps(fiz1,tz);
1269
1270             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1271             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1272             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1273             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1274             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1275
1276             }
1277
1278             /**************************
1279              * CALCULATE INTERACTIONS *
1280              **************************/
1281
1282             if (gmx_mm_any_lt(rsq20,rcutoff2))
1283             {
1284
1285             /* Compute parameters for interactions between i and j atoms */
1286             qq20             = _mm_mul_ps(iq2,jq0);
1287
1288             /* REACTION-FIELD ELECTROSTATICS */
1289             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1290
1291             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1292
1293             fscal            = felec;
1294
1295             fscal            = _mm_and_ps(fscal,cutoff_mask);
1296
1297             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1298
1299             /* Calculate temporary vectorial force */
1300             tx               = _mm_mul_ps(fscal,dx20);
1301             ty               = _mm_mul_ps(fscal,dy20);
1302             tz               = _mm_mul_ps(fscal,dz20);
1303
1304             /* Update vectorial force */
1305             fix2             = _mm_add_ps(fix2,tx);
1306             fiy2             = _mm_add_ps(fiy2,ty);
1307             fiz2             = _mm_add_ps(fiz2,tz);
1308
1309             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1310             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1311             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1312             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1313             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1314
1315             }
1316
1317             /**************************
1318              * CALCULATE INTERACTIONS *
1319              **************************/
1320
1321             if (gmx_mm_any_lt(rsq30,rcutoff2))
1322             {
1323
1324             /* Compute parameters for interactions between i and j atoms */
1325             qq30             = _mm_mul_ps(iq3,jq0);
1326
1327             /* REACTION-FIELD ELECTROSTATICS */
1328             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1329
1330             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1331
1332             fscal            = felec;
1333
1334             fscal            = _mm_and_ps(fscal,cutoff_mask);
1335
1336             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1337
1338             /* Calculate temporary vectorial force */
1339             tx               = _mm_mul_ps(fscal,dx30);
1340             ty               = _mm_mul_ps(fscal,dy30);
1341             tz               = _mm_mul_ps(fscal,dz30);
1342
1343             /* Update vectorial force */
1344             fix3             = _mm_add_ps(fix3,tx);
1345             fiy3             = _mm_add_ps(fiy3,ty);
1346             fiz3             = _mm_add_ps(fiz3,tz);
1347
1348             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1349             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1350             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1351             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1352             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1353
1354             }
1355
1356             /* Inner loop uses 139 flops */
1357         }
1358
1359         /* End of innermost loop */
1360
1361         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1362                                               f+i_coord_offset,fshift+i_shift_offset);
1363
1364         /* Increment number of inner iterations */
1365         inneriter                  += j_index_end - j_index_start;
1366
1367         /* Outer loop uses 24 flops */
1368     }
1369
1370     /* Increment number of outer iterations */
1371     outeriter        += nri;
1372
1373     /* Update outer/inner flops */
1374
1375     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*139);
1376 }