src/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             fjx0             = _mm_setzero_ps();
 230             fjy0             = _mm_setzero_ps();
 231             fjz0             = _mm_setzero_ps();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 239                                          vdwparam+vdwioffset0+vdwjidx0B,
 240                                          vdwparam+vdwioffset0+vdwjidx0C,
 241                                          vdwparam+vdwioffset0+vdwjidx0D,
 242                                          &c6_00,&c12_00);
 243
 244             /* LENNARD-JONES DISPERSION/REPULSION */
 245
 246             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 247             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 248             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 249             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 250             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 254
 255             fscal            = fvdw;
 256
 257             /* Calculate temporary vectorial force */
 258             tx               = _mm_mul_ps(fscal,dx00);
 259             ty               = _mm_mul_ps(fscal,dy00);
 260             tz               = _mm_mul_ps(fscal,dz00);
 261
 262             /* Update vectorial force */
 263             fix0             = _mm_add_ps(fix0,tx);
 264             fiy0             = _mm_add_ps(fiy0,ty);
 265             fiz0             = _mm_add_ps(fiz0,tz);
 266
 267             fjx0             = _mm_add_ps(fjx0,tx);
 268             fjy0             = _mm_add_ps(fjy0,ty);
 269             fjz0             = _mm_add_ps(fjz0,tz);
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             r10              = _mm_mul_ps(rsq10,rinv10);
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             qq10             = _mm_mul_ps(iq1,jq0);
 279
 280             /* Calculate table index by multiplying r with table scale and truncate to integer */
 281             rt               = _mm_mul_ps(r10,vftabscale);
 282             vfitab           = _mm_cvttps_epi32(rt);
 283             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 284             vfitab           = _mm_slli_epi32(vfitab,2);
 285
 286             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 287             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 288             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 289             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 290             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 291             _MM_TRANSPOSE4_PS(Y,F,G,H);
 292             Heps             = _mm_mul_ps(vfeps,H);
 293             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 294             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 295             velec            = _mm_mul_ps(qq10,VV);
 296             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 297             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velecsum         = _mm_add_ps(velecsum,velec);
 301
 302             fscal            = felec;
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm_mul_ps(fscal,dx10);
 306             ty               = _mm_mul_ps(fscal,dy10);
 307             tz               = _mm_mul_ps(fscal,dz10);
 308
 309             /* Update vectorial force */
 310             fix1             = _mm_add_ps(fix1,tx);
 311             fiy1             = _mm_add_ps(fiy1,ty);
 312             fiz1             = _mm_add_ps(fiz1,tz);
 313
 314             fjx0             = _mm_add_ps(fjx0,tx);
 315             fjy0             = _mm_add_ps(fjy0,ty);
 316             fjz0             = _mm_add_ps(fjz0,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             fjx0             = _mm_add_ps(fjx0,tx);
 362             fjy0             = _mm_add_ps(fjy0,ty);
 363             fjz0             = _mm_add_ps(fjz0,tz);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             r30              = _mm_mul_ps(rsq30,rinv30);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq30             = _mm_mul_ps(iq3,jq0);
 373
 374             /* Calculate table index by multiplying r with table scale and truncate to integer */
 375             rt               = _mm_mul_ps(r30,vftabscale);
 376             vfitab           = _mm_cvttps_epi32(rt);
 377             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 378             vfitab           = _mm_slli_epi32(vfitab,2);
 379
 380             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 381             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 382             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 383             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 384             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 385             _MM_TRANSPOSE4_PS(Y,F,G,H);
 386             Heps             = _mm_mul_ps(vfeps,H);
 387             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 388             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 389             velec            = _mm_mul_ps(qq30,VV);
 390             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 391             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velecsum         = _mm_add_ps(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm_mul_ps(fscal,dx30);
 400             ty               = _mm_mul_ps(fscal,dy30);
 401             tz               = _mm_mul_ps(fscal,dz30);
 402
 403             /* Update vectorial force */
 404             fix3             = _mm_add_ps(fix3,tx);
 405             fiy3             = _mm_add_ps(fiy3,ty);
 406             fiz3             = _mm_add_ps(fiz3,tz);
 407
 408             fjx0             = _mm_add_ps(fjx0,tx);
 409             fjy0             = _mm_add_ps(fjy0,ty);
 410             fjz0             = _mm_add_ps(fjz0,tz);
 411
 412             fjptrA             = f+j_coord_offsetA;
 413             fjptrB             = f+j_coord_offsetB;
 414             fjptrC             = f+j_coord_offsetC;
 415             fjptrD             = f+j_coord_offsetD;
 416
 417             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 418
 419             /* Inner loop uses 161 flops */
 420         }
 421
 422         if(jidx<j_index_end)
 423         {
 424
 425             /* Get j neighbor index, and coordinate index */
 426             jnrlistA         = jjnr[jidx];
 427             jnrlistB         = jjnr[jidx+1];
 428             jnrlistC         = jjnr[jidx+2];
 429             jnrlistD         = jjnr[jidx+3];
 430             /* Sign of each element will be negative for non-real atoms.
 431              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 432              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 433              */
 434             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 435             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 436             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 437             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 438             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 439             j_coord_offsetA  = DIM*jnrA;
 440             j_coord_offsetB  = DIM*jnrB;
 441             j_coord_offsetC  = DIM*jnrC;
 442             j_coord_offsetD  = DIM*jnrD;
 443
 444             /* load j atom coordinates */
 445             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 446                                               x+j_coord_offsetC,x+j_coord_offsetD,
 447                                               &jx0,&jy0,&jz0);
 448
 449             /* Calculate displacement vector */
 450             dx00             = _mm_sub_ps(ix0,jx0);
 451             dy00             = _mm_sub_ps(iy0,jy0);
 452             dz00             = _mm_sub_ps(iz0,jz0);
 453             dx10             = _mm_sub_ps(ix1,jx0);
 454             dy10             = _mm_sub_ps(iy1,jy0);
 455             dz10             = _mm_sub_ps(iz1,jz0);
 456             dx20             = _mm_sub_ps(ix2,jx0);
 457             dy20             = _mm_sub_ps(iy2,jy0);
 458             dz20             = _mm_sub_ps(iz2,jz0);
 459             dx30             = _mm_sub_ps(ix3,jx0);
 460             dy30             = _mm_sub_ps(iy3,jy0);
 461             dz30             = _mm_sub_ps(iz3,jz0);
 462
 463             /* Calculate squared distance and things based on it */
 464             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 465             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 466             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 467             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 468
 469             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 470             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 471             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 472
 473             rinvsq00         = gmx_mm_inv_ps(rsq00);
 474
 475             /* Load parameters for j particles */
 476             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 477                                                               charge+jnrC+0,charge+jnrD+0);
 478             vdwjidx0A        = 2*vdwtype[jnrA+0];
 479             vdwjidx0B        = 2*vdwtype[jnrB+0];
 480             vdwjidx0C        = 2*vdwtype[jnrC+0];
 481             vdwjidx0D        = 2*vdwtype[jnrD+0];
 482
 483             fjx0             = _mm_setzero_ps();
 484             fjy0             = _mm_setzero_ps();
 485             fjz0             = _mm_setzero_ps();
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             /* Compute parameters for interactions between i and j atoms */
 492             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 493                                          vdwparam+vdwioffset0+vdwjidx0B,
 494                                          vdwparam+vdwioffset0+vdwjidx0C,
 495                                          vdwparam+vdwioffset0+vdwjidx0D,
 496                                          &c6_00,&c12_00);
 497
 498             /* LENNARD-JONES DISPERSION/REPULSION */
 499
 500             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 501             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 502             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 503             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 504             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 505
 506             /* Update potential sum for this i atom from the interaction with this j atom. */
 507             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 508             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 509
 510             fscal            = fvdw;
 511
 512             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 513
 514             /* Calculate temporary vectorial force */
 515             tx               = _mm_mul_ps(fscal,dx00);
 516             ty               = _mm_mul_ps(fscal,dy00);
 517             tz               = _mm_mul_ps(fscal,dz00);
 518
 519             /* Update vectorial force */
 520             fix0             = _mm_add_ps(fix0,tx);
 521             fiy0             = _mm_add_ps(fiy0,ty);
 522             fiz0             = _mm_add_ps(fiz0,tz);
 523
 524             fjx0             = _mm_add_ps(fjx0,tx);
 525             fjy0             = _mm_add_ps(fjy0,ty);
 526             fjz0             = _mm_add_ps(fjz0,tz);
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             r10              = _mm_mul_ps(rsq10,rinv10);
 533             r10              = _mm_andnot_ps(dummy_mask,r10);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq10             = _mm_mul_ps(iq1,jq0);
 537
 538             /* Calculate table index by multiplying r with table scale and truncate to integer */
 539             rt               = _mm_mul_ps(r10,vftabscale);
 540             vfitab           = _mm_cvttps_epi32(rt);
 541             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 542             vfitab           = _mm_slli_epi32(vfitab,2);
 543
 544             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 545             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 546             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 547             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 548             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 549             _MM_TRANSPOSE4_PS(Y,F,G,H);
 550             Heps             = _mm_mul_ps(vfeps,H);
 551             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 552             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 553             velec            = _mm_mul_ps(qq10,VV);
 554             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 555             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velec            = _mm_andnot_ps(dummy_mask,velec);
 559             velecsum         = _mm_add_ps(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm_mul_ps(fscal,dx10);
 567             ty               = _mm_mul_ps(fscal,dy10);
 568             tz               = _mm_mul_ps(fscal,dz10);
 569
 570             /* Update vectorial force */
 571             fix1             = _mm_add_ps(fix1,tx);
 572             fiy1             = _mm_add_ps(fiy1,ty);
 573             fiz1             = _mm_add_ps(fiz1,tz);
 574
 575             fjx0             = _mm_add_ps(fjx0,tx);
 576             fjy0             = _mm_add_ps(fjy0,ty);
 577             fjz0             = _mm_add_ps(fjz0,tz);
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             r20              = _mm_mul_ps(rsq20,rinv20);
 584             r20              = _mm_andnot_ps(dummy_mask,r20);
 585
 586             /* Compute parameters for interactions between i and j atoms */
 587             qq20             = _mm_mul_ps(iq2,jq0);
 588
 589             /* Calculate table index by multiplying r with table scale and truncate to integer */
 590             rt               = _mm_mul_ps(r20,vftabscale);
 591             vfitab           = _mm_cvttps_epi32(rt);
 592             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 593             vfitab           = _mm_slli_epi32(vfitab,2);
 594
 595             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 596             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 597             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 598             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 599             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 600             _MM_TRANSPOSE4_PS(Y,F,G,H);
 601             Heps             = _mm_mul_ps(vfeps,H);
 602             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 603             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 604             velec            = _mm_mul_ps(qq20,VV);
 605             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 606             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 607
 608             /* Update potential sum for this i atom from the interaction with this j atom. */
 609             velec            = _mm_andnot_ps(dummy_mask,velec);
 610             velecsum         = _mm_add_ps(velecsum,velec);
 611
 612             fscal            = felec;
 613
 614             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 615
 616             /* Calculate temporary vectorial force */
 617             tx               = _mm_mul_ps(fscal,dx20);
 618             ty               = _mm_mul_ps(fscal,dy20);
 619             tz               = _mm_mul_ps(fscal,dz20);
 620
 621             /* Update vectorial force */
 622             fix2             = _mm_add_ps(fix2,tx);
 623             fiy2             = _mm_add_ps(fiy2,ty);
 624             fiz2             = _mm_add_ps(fiz2,tz);
 625
 626             fjx0             = _mm_add_ps(fjx0,tx);
 627             fjy0             = _mm_add_ps(fjy0,ty);
 628             fjz0             = _mm_add_ps(fjz0,tz);
 629
 630             /**************************
 631              * CALCULATE INTERACTIONS *
 632              **************************/
 633
 634             r30              = _mm_mul_ps(rsq30,rinv30);
 635             r30              = _mm_andnot_ps(dummy_mask,r30);
 636
 637             /* Compute parameters for interactions between i and j atoms */
 638             qq30             = _mm_mul_ps(iq3,jq0);
 639
 640             /* Calculate table index by multiplying r with table scale and truncate to integer */
 641             rt               = _mm_mul_ps(r30,vftabscale);
 642             vfitab           = _mm_cvttps_epi32(rt);
 643             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 644             vfitab           = _mm_slli_epi32(vfitab,2);
 645
 646             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 647             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 648             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 649             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 650             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 651             _MM_TRANSPOSE4_PS(Y,F,G,H);
 652             Heps             = _mm_mul_ps(vfeps,H);
 653             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 654             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 655             velec            = _mm_mul_ps(qq30,VV);
 656             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 657             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 658
 659             /* Update potential sum for this i atom from the interaction with this j atom. */
 660             velec            = _mm_andnot_ps(dummy_mask,velec);
 661             velecsum         = _mm_add_ps(velecsum,velec);
 662
 663             fscal            = felec;
 664
 665             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 666
 667             /* Calculate temporary vectorial force */
 668             tx               = _mm_mul_ps(fscal,dx30);
 669             ty               = _mm_mul_ps(fscal,dy30);
 670             tz               = _mm_mul_ps(fscal,dz30);
 671
 672             /* Update vectorial force */
 673             fix3             = _mm_add_ps(fix3,tx);
 674             fiy3             = _mm_add_ps(fiy3,ty);
 675             fiz3             = _mm_add_ps(fiz3,tz);
 676
 677             fjx0             = _mm_add_ps(fjx0,tx);
 678             fjy0             = _mm_add_ps(fjy0,ty);
 679             fjz0             = _mm_add_ps(fjz0,tz);
 680
 681             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 682             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 683             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 684             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 685
 686             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 687
 688             /* Inner loop uses 164 flops */
 689         }
 690
 691         /* End of innermost loop */
 692
 693         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 694                                               f+i_coord_offset,fshift+i_shift_offset);
 695
 696         ggid                        = gid[iidx];
 697         /* Update potential energies */
 698         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 699         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 700
 701         /* Increment number of inner iterations */
 702         inneriter                  += j_index_end - j_index_start;
 703
 704         /* Outer loop uses 26 flops */
 705     }
 706
 707     /* Increment number of outer iterations */
 708     outeriter        += nri;
 709
 710     /* Update outer/inner flops */
 711
 712     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*164);
 713 }
 714 /*
 715  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_single
 716  * Electrostatics interaction: CubicSplineTable
 717  * VdW interaction:            LennardJones
 718  * Geometry:                   Water4-Particle
 719  * Calculate force/pot:        Force
 720  */
 721 void
 722 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_single
 723                     (t_nblist * gmx_restrict                nlist,
 724                      rvec * gmx_restrict                    xx,
 725                      rvec * gmx_restrict                    ff,
 726                      t_forcerec * gmx_restrict              fr,
 727                      t_mdatoms * gmx_restrict               mdatoms,
 728                      nb_kernel_data_t * gmx_restrict        kernel_data,
 729                      t_nrnb * gmx_restrict                  nrnb)
 730 {
 731     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 732      * just 0 for non-waters.
 733      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 734      * jnr indices corresponding to data put in the four positions in the SIMD register.
 735      */
 736     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 737     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 738     int              jnrA,jnrB,jnrC,jnrD;
 739     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 740     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 741     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 742     real             rcutoff_scalar;
 743     real             *shiftvec,*fshift,*x,*f;
 744     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 745     real             scratch[4*DIM];
 746     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 747     int              vdwioffset0;
 748     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 749     int              vdwioffset1;
 750     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 751     int              vdwioffset2;
 752     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 753     int              vdwioffset3;
 754     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 755     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 756     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 757     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 758     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 759     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 760     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 761     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 762     real             *charge;
 763     int              nvdwtype;
 764     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 765     int              *vdwtype;
 766     real             *vdwparam;
 767     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 768     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 769     __m128i          vfitab;
 770     __m128i          ifour       = _mm_set1_epi32(4);
 771     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 772     real             *vftab;
 773     __m128           dummy_mask,cutoff_mask;
 774     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 775     __m128           one     = _mm_set1_ps(1.0);
 776     __m128           two     = _mm_set1_ps(2.0);
 777     x                = xx[0];
 778     f                = ff[0];
 779
 780     nri              = nlist->nri;
 781     iinr             = nlist->iinr;
 782     jindex           = nlist->jindex;
 783     jjnr             = nlist->jjnr;
 784     shiftidx         = nlist->shift;
 785     gid              = nlist->gid;
 786     shiftvec         = fr->shift_vec[0];
 787     fshift           = fr->fshift[0];
 788     facel            = _mm_set1_ps(fr->epsfac);
 789     charge           = mdatoms->chargeA;
 790     nvdwtype         = fr->ntype;
 791     vdwparam         = fr->nbfp;
 792     vdwtype          = mdatoms->typeA;
 793
 794     vftab            = kernel_data->table_elec->data;
 795     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 796
 797     /* Setup water-specific parameters */
 798     inr              = nlist->iinr[0];
 799     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 800     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 801     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 802     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 803
 804     /* Avoid stupid compiler warnings */
 805     jnrA = jnrB = jnrC = jnrD = 0;
 806     j_coord_offsetA = 0;
 807     j_coord_offsetB = 0;
 808     j_coord_offsetC = 0;
 809     j_coord_offsetD = 0;
 810
 811     outeriter        = 0;
 812     inneriter        = 0;
 813
 814     for(iidx=0;iidx<4*DIM;iidx++)
 815     {
 816         scratch[iidx] = 0.0;
 817     }
 818
 819     /* Start outer loop over neighborlists */
 820     for(iidx=0; iidx<nri; iidx++)
 821     {
 822         /* Load shift vector for this list */
 823         i_shift_offset   = DIM*shiftidx[iidx];
 824
 825         /* Load limits for loop over neighbors */
 826         j_index_start    = jindex[iidx];
 827         j_index_end      = jindex[iidx+1];
 828
 829         /* Get outer coordinate index */
 830         inr              = iinr[iidx];
 831         i_coord_offset   = DIM*inr;
 832
 833         /* Load i particle coords and add shift vector */
 834         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 835                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 836
 837         fix0             = _mm_setzero_ps();
 838         fiy0             = _mm_setzero_ps();
 839         fiz0             = _mm_setzero_ps();
 840         fix1             = _mm_setzero_ps();
 841         fiy1             = _mm_setzero_ps();
 842         fiz1             = _mm_setzero_ps();
 843         fix2             = _mm_setzero_ps();
 844         fiy2             = _mm_setzero_ps();
 845         fiz2             = _mm_setzero_ps();
 846         fix3             = _mm_setzero_ps();
 847         fiy3             = _mm_setzero_ps();
 848         fiz3             = _mm_setzero_ps();
 849
 850         /* Start inner kernel loop */
 851         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 852         {
 853
 854             /* Get j neighbor index, and coordinate index */
 855             jnrA             = jjnr[jidx];
 856             jnrB             = jjnr[jidx+1];
 857             jnrC             = jjnr[jidx+2];
 858             jnrD             = jjnr[jidx+3];
 859             j_coord_offsetA  = DIM*jnrA;
 860             j_coord_offsetB  = DIM*jnrB;
 861             j_coord_offsetC  = DIM*jnrC;
 862             j_coord_offsetD  = DIM*jnrD;
 863
 864             /* load j atom coordinates */
 865             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 866                                               x+j_coord_offsetC,x+j_coord_offsetD,
 867                                               &jx0,&jy0,&jz0);
 868
 869             /* Calculate displacement vector */
 870             dx00             = _mm_sub_ps(ix0,jx0);
 871             dy00             = _mm_sub_ps(iy0,jy0);
 872             dz00             = _mm_sub_ps(iz0,jz0);
 873             dx10             = _mm_sub_ps(ix1,jx0);
 874             dy10             = _mm_sub_ps(iy1,jy0);
 875             dz10             = _mm_sub_ps(iz1,jz0);
 876             dx20             = _mm_sub_ps(ix2,jx0);
 877             dy20             = _mm_sub_ps(iy2,jy0);
 878             dz20             = _mm_sub_ps(iz2,jz0);
 879             dx30             = _mm_sub_ps(ix3,jx0);
 880             dy30             = _mm_sub_ps(iy3,jy0);
 881             dz30             = _mm_sub_ps(iz3,jz0);
 882
 883             /* Calculate squared distance and things based on it */
 884             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 885             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 886             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 887             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 888
 889             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 890             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 891             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 892
 893             rinvsq00         = gmx_mm_inv_ps(rsq00);
 894
 895             /* Load parameters for j particles */
 896             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 897                                                               charge+jnrC+0,charge+jnrD+0);
 898             vdwjidx0A        = 2*vdwtype[jnrA+0];
 899             vdwjidx0B        = 2*vdwtype[jnrB+0];
 900             vdwjidx0C        = 2*vdwtype[jnrC+0];
 901             vdwjidx0D        = 2*vdwtype[jnrD+0];
 902
 903             fjx0             = _mm_setzero_ps();
 904             fjy0             = _mm_setzero_ps();
 905             fjz0             = _mm_setzero_ps();
 906
 907             /**************************
 908              * CALCULATE INTERACTIONS *
 909              **************************/
 910
 911             /* Compute parameters for interactions between i and j atoms */
 912             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 913                                          vdwparam+vdwioffset0+vdwjidx0B,
 914                                          vdwparam+vdwioffset0+vdwjidx0C,
 915                                          vdwparam+vdwioffset0+vdwjidx0D,
 916                                          &c6_00,&c12_00);
 917
 918             /* LENNARD-JONES DISPERSION/REPULSION */
 919
 920             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 921             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 922
 923             fscal            = fvdw;
 924
 925             /* Calculate temporary vectorial force */
 926             tx               = _mm_mul_ps(fscal,dx00);
 927             ty               = _mm_mul_ps(fscal,dy00);
 928             tz               = _mm_mul_ps(fscal,dz00);
 929
 930             /* Update vectorial force */
 931             fix0             = _mm_add_ps(fix0,tx);
 932             fiy0             = _mm_add_ps(fiy0,ty);
 933             fiz0             = _mm_add_ps(fiz0,tz);
 934
 935             fjx0             = _mm_add_ps(fjx0,tx);
 936             fjy0             = _mm_add_ps(fjy0,ty);
 937             fjz0             = _mm_add_ps(fjz0,tz);
 938
 939             /**************************
 940              * CALCULATE INTERACTIONS *
 941              **************************/
 942
 943             r10              = _mm_mul_ps(rsq10,rinv10);
 944
 945             /* Compute parameters for interactions between i and j atoms */
 946             qq10             = _mm_mul_ps(iq1,jq0);
 947
 948             /* Calculate table index by multiplying r with table scale and truncate to integer */
 949             rt               = _mm_mul_ps(r10,vftabscale);
 950             vfitab           = _mm_cvttps_epi32(rt);
 951             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 952             vfitab           = _mm_slli_epi32(vfitab,2);
 953
 954             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 955             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 956             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 957             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 958             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 959             _MM_TRANSPOSE4_PS(Y,F,G,H);
 960             Heps             = _mm_mul_ps(vfeps,H);
 961             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 962             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 963             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 964
 965             fscal            = felec;
 966
 967             /* Calculate temporary vectorial force */
 968             tx               = _mm_mul_ps(fscal,dx10);
 969             ty               = _mm_mul_ps(fscal,dy10);
 970             tz               = _mm_mul_ps(fscal,dz10);
 971
 972             /* Update vectorial force */
 973             fix1             = _mm_add_ps(fix1,tx);
 974             fiy1             = _mm_add_ps(fiy1,ty);
 975             fiz1             = _mm_add_ps(fiz1,tz);
 976
 977             fjx0             = _mm_add_ps(fjx0,tx);
 978             fjy0             = _mm_add_ps(fjy0,ty);
 979             fjz0             = _mm_add_ps(fjz0,tz);
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             r20              = _mm_mul_ps(rsq20,rinv20);
 986
 987             /* Compute parameters for interactions between i and j atoms */
 988             qq20             = _mm_mul_ps(iq2,jq0);
 989
 990             /* Calculate table index by multiplying r with table scale and truncate to integer */
 991             rt               = _mm_mul_ps(r20,vftabscale);
 992             vfitab           = _mm_cvttps_epi32(rt);
 993             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 994             vfitab           = _mm_slli_epi32(vfitab,2);
 995
 996             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 997             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 998             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 999             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1000             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1001             _MM_TRANSPOSE4_PS(Y,F,G,H);
1002             Heps             = _mm_mul_ps(vfeps,H);
1003             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1004             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1005             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1006
1007             fscal            = felec;
1008
1009             /* Calculate temporary vectorial force */
1010             tx               = _mm_mul_ps(fscal,dx20);
1011             ty               = _mm_mul_ps(fscal,dy20);
1012             tz               = _mm_mul_ps(fscal,dz20);
1013
1014             /* Update vectorial force */
1015             fix2             = _mm_add_ps(fix2,tx);
1016             fiy2             = _mm_add_ps(fiy2,ty);
1017             fiz2             = _mm_add_ps(fiz2,tz);
1018
1019             fjx0             = _mm_add_ps(fjx0,tx);
1020             fjy0             = _mm_add_ps(fjy0,ty);
1021             fjz0             = _mm_add_ps(fjz0,tz);
1022
1023             /**************************
1024              * CALCULATE INTERACTIONS *
1025              **************************/
1026
1027             r30              = _mm_mul_ps(rsq30,rinv30);
1028
1029             /* Compute parameters for interactions between i and j atoms */
1030             qq30             = _mm_mul_ps(iq3,jq0);
1031
1032             /* Calculate table index by multiplying r with table scale and truncate to integer */
1033             rt               = _mm_mul_ps(r30,vftabscale);
1034             vfitab           = _mm_cvttps_epi32(rt);
1035             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1036             vfitab           = _mm_slli_epi32(vfitab,2);
1037
1038             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1039             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1040             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1041             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1042             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1043             _MM_TRANSPOSE4_PS(Y,F,G,H);
1044             Heps             = _mm_mul_ps(vfeps,H);
1045             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1046             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1047             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1048
1049             fscal            = felec;
1050
1051             /* Calculate temporary vectorial force */
1052             tx               = _mm_mul_ps(fscal,dx30);
1053             ty               = _mm_mul_ps(fscal,dy30);
1054             tz               = _mm_mul_ps(fscal,dz30);
1055
1056             /* Update vectorial force */
1057             fix3             = _mm_add_ps(fix3,tx);
1058             fiy3             = _mm_add_ps(fiy3,ty);
1059             fiz3             = _mm_add_ps(fiz3,tz);
1060
1061             fjx0             = _mm_add_ps(fjx0,tx);
1062             fjy0             = _mm_add_ps(fjy0,ty);
1063             fjz0             = _mm_add_ps(fjz0,tz);
1064
1065             fjptrA             = f+j_coord_offsetA;
1066             fjptrB             = f+j_coord_offsetB;
1067             fjptrC             = f+j_coord_offsetC;
1068             fjptrD             = f+j_coord_offsetD;
1069
1070             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1071
1072             /* Inner loop uses 144 flops */
1073         }
1074
1075         if(jidx<j_index_end)
1076         {
1077
1078             /* Get j neighbor index, and coordinate index */
1079             jnrlistA         = jjnr[jidx];
1080             jnrlistB         = jjnr[jidx+1];
1081             jnrlistC         = jjnr[jidx+2];
1082             jnrlistD         = jjnr[jidx+3];
1083             /* Sign of each element will be negative for non-real atoms.
1084              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1085              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1086              */
1087             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1088             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1089             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1090             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1091             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1092             j_coord_offsetA  = DIM*jnrA;
1093             j_coord_offsetB  = DIM*jnrB;
1094             j_coord_offsetC  = DIM*jnrC;
1095             j_coord_offsetD  = DIM*jnrD;
1096
1097             /* load j atom coordinates */
1098             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1099                                               x+j_coord_offsetC,x+j_coord_offsetD,
1100                                               &jx0,&jy0,&jz0);
1101
1102             /* Calculate displacement vector */
1103             dx00             = _mm_sub_ps(ix0,jx0);
1104             dy00             = _mm_sub_ps(iy0,jy0);
1105             dz00             = _mm_sub_ps(iz0,jz0);
1106             dx10             = _mm_sub_ps(ix1,jx0);
1107             dy10             = _mm_sub_ps(iy1,jy0);
1108             dz10             = _mm_sub_ps(iz1,jz0);
1109             dx20             = _mm_sub_ps(ix2,jx0);
1110             dy20             = _mm_sub_ps(iy2,jy0);
1111             dz20             = _mm_sub_ps(iz2,jz0);
1112             dx30             = _mm_sub_ps(ix3,jx0);
1113             dy30             = _mm_sub_ps(iy3,jy0);
1114             dz30             = _mm_sub_ps(iz3,jz0);
1115
1116             /* Calculate squared distance and things based on it */
1117             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1118             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1119             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1120             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1121
1122             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1123             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1124             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1125
1126             rinvsq00         = gmx_mm_inv_ps(rsq00);
1127
1128             /* Load parameters for j particles */
1129             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1130                                                               charge+jnrC+0,charge+jnrD+0);
1131             vdwjidx0A        = 2*vdwtype[jnrA+0];
1132             vdwjidx0B        = 2*vdwtype[jnrB+0];
1133             vdwjidx0C        = 2*vdwtype[jnrC+0];
1134             vdwjidx0D        = 2*vdwtype[jnrD+0];
1135
1136             fjx0             = _mm_setzero_ps();
1137             fjy0             = _mm_setzero_ps();
1138             fjz0             = _mm_setzero_ps();
1139
1140             /**************************
1141              * CALCULATE INTERACTIONS *
1142              **************************/
1143
1144             /* Compute parameters for interactions between i and j atoms */
1145             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1146                                          vdwparam+vdwioffset0+vdwjidx0B,
1147                                          vdwparam+vdwioffset0+vdwjidx0C,
1148                                          vdwparam+vdwioffset0+vdwjidx0D,
1149                                          &c6_00,&c12_00);
1150
1151             /* LENNARD-JONES DISPERSION/REPULSION */
1152
1153             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1154             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1155
1156             fscal            = fvdw;
1157
1158             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1159
1160             /* Calculate temporary vectorial force */
1161             tx               = _mm_mul_ps(fscal,dx00);
1162             ty               = _mm_mul_ps(fscal,dy00);
1163             tz               = _mm_mul_ps(fscal,dz00);
1164
1165             /* Update vectorial force */
1166             fix0             = _mm_add_ps(fix0,tx);
1167             fiy0             = _mm_add_ps(fiy0,ty);
1168             fiz0             = _mm_add_ps(fiz0,tz);
1169
1170             fjx0             = _mm_add_ps(fjx0,tx);
1171             fjy0             = _mm_add_ps(fjy0,ty);
1172             fjz0             = _mm_add_ps(fjz0,tz);
1173
1174             /**************************
1175              * CALCULATE INTERACTIONS *
1176              **************************/
1177
1178             r10              = _mm_mul_ps(rsq10,rinv10);
1179             r10              = _mm_andnot_ps(dummy_mask,r10);
1180
1181             /* Compute parameters for interactions between i and j atoms */
1182             qq10             = _mm_mul_ps(iq1,jq0);
1183
1184             /* Calculate table index by multiplying r with table scale and truncate to integer */
1185             rt               = _mm_mul_ps(r10,vftabscale);
1186             vfitab           = _mm_cvttps_epi32(rt);
1187             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1188             vfitab           = _mm_slli_epi32(vfitab,2);
1189
1190             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1191             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1192             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1193             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1194             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1195             _MM_TRANSPOSE4_PS(Y,F,G,H);
1196             Heps             = _mm_mul_ps(vfeps,H);
1197             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1198             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1199             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1200
1201             fscal            = felec;
1202
1203             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1204
1205             /* Calculate temporary vectorial force */
1206             tx               = _mm_mul_ps(fscal,dx10);
1207             ty               = _mm_mul_ps(fscal,dy10);
1208             tz               = _mm_mul_ps(fscal,dz10);
1209
1210             /* Update vectorial force */
1211             fix1             = _mm_add_ps(fix1,tx);
1212             fiy1             = _mm_add_ps(fiy1,ty);
1213             fiz1             = _mm_add_ps(fiz1,tz);
1214
1215             fjx0             = _mm_add_ps(fjx0,tx);
1216             fjy0             = _mm_add_ps(fjy0,ty);
1217             fjz0             = _mm_add_ps(fjz0,tz);
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             r20              = _mm_mul_ps(rsq20,rinv20);
1224             r20              = _mm_andnot_ps(dummy_mask,r20);
1225
1226             /* Compute parameters for interactions between i and j atoms */
1227             qq20             = _mm_mul_ps(iq2,jq0);
1228
1229             /* Calculate table index by multiplying r with table scale and truncate to integer */
1230             rt               = _mm_mul_ps(r20,vftabscale);
1231             vfitab           = _mm_cvttps_epi32(rt);
1232             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1233             vfitab           = _mm_slli_epi32(vfitab,2);
1234
1235             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1236             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1237             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1238             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1239             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1240             _MM_TRANSPOSE4_PS(Y,F,G,H);
1241             Heps             = _mm_mul_ps(vfeps,H);
1242             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1243             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1244             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1245
1246             fscal            = felec;
1247
1248             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1249
1250             /* Calculate temporary vectorial force */
1251             tx               = _mm_mul_ps(fscal,dx20);
1252             ty               = _mm_mul_ps(fscal,dy20);
1253             tz               = _mm_mul_ps(fscal,dz20);
1254
1255             /* Update vectorial force */
1256             fix2             = _mm_add_ps(fix2,tx);
1257             fiy2             = _mm_add_ps(fiy2,ty);
1258             fiz2             = _mm_add_ps(fiz2,tz);
1259
1260             fjx0             = _mm_add_ps(fjx0,tx);
1261             fjy0             = _mm_add_ps(fjy0,ty);
1262             fjz0             = _mm_add_ps(fjz0,tz);
1263
1264             /**************************
1265              * CALCULATE INTERACTIONS *
1266              **************************/
1267
1268             r30              = _mm_mul_ps(rsq30,rinv30);
1269             r30              = _mm_andnot_ps(dummy_mask,r30);
1270
1271             /* Compute parameters for interactions between i and j atoms */
1272             qq30             = _mm_mul_ps(iq3,jq0);
1273
1274             /* Calculate table index by multiplying r with table scale and truncate to integer */
1275             rt               = _mm_mul_ps(r30,vftabscale);
1276             vfitab           = _mm_cvttps_epi32(rt);
1277             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1278             vfitab           = _mm_slli_epi32(vfitab,2);
1279
1280             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1281             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1282             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1283             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1284             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1285             _MM_TRANSPOSE4_PS(Y,F,G,H);
1286             Heps             = _mm_mul_ps(vfeps,H);
1287             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1288             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1289             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1290
1291             fscal            = felec;
1292
1293             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1294
1295             /* Calculate temporary vectorial force */
1296             tx               = _mm_mul_ps(fscal,dx30);
1297             ty               = _mm_mul_ps(fscal,dy30);
1298             tz               = _mm_mul_ps(fscal,dz30);
1299
1300             /* Update vectorial force */
1301             fix3             = _mm_add_ps(fix3,tx);
1302             fiy3             = _mm_add_ps(fiy3,ty);
1303             fiz3             = _mm_add_ps(fiz3,tz);
1304
1305             fjx0             = _mm_add_ps(fjx0,tx);
1306             fjy0             = _mm_add_ps(fjy0,ty);
1307             fjz0             = _mm_add_ps(fjz0,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
1314             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1315
1316             /* Inner loop uses 147 flops */
1317         }
1318
1319         /* End of innermost loop */
1320
1321         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1322                                               f+i_coord_offset,fshift+i_shift_offset);
1323
1324         /* Increment number of inner iterations */
1325         inneriter                  += j_index_end - j_index_start;
1326
1327         /* Outer loop uses 24 flops */
1328     }
1329
1330     /* Increment number of outer iterations */
1331     outeriter        += nri;
1332
1333     /* Update outer/inner flops */
1334
1335     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1336 }