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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwioffset3;
  73     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  75     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_elec->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 119     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 120     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138         shX              = shiftvec[i_shift_offset+XX];
 139         shY              = shiftvec[i_shift_offset+YY];
 140         shZ              = shiftvec[i_shift_offset+ZZ];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 152         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 153         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 154         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 155         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 156         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 157         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 158         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 159         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 160         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 161         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 162         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 163
 164         fix0             = _mm_setzero_ps();
 165         fiy0             = _mm_setzero_ps();
 166         fiz0             = _mm_setzero_ps();
 167         fix1             = _mm_setzero_ps();
 168         fiy1             = _mm_setzero_ps();
 169         fiz1             = _mm_setzero_ps();
 170         fix2             = _mm_setzero_ps();
 171         fiy2             = _mm_setzero_ps();
 172         fiz2             = _mm_setzero_ps();
 173         fix3             = _mm_setzero_ps();
 174         fiy3             = _mm_setzero_ps();
 175         fiz3             = _mm_setzero_ps();
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm_setzero_ps();
 179         vvdwsum          = _mm_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               x+j_coord_offsetC,x+j_coord_offsetD,
 199                                               &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm_sub_ps(ix0,jx0);
 203             dy00             = _mm_sub_ps(iy0,jy0);
 204             dz00             = _mm_sub_ps(iz0,jz0);
 205             dx10             = _mm_sub_ps(ix1,jx0);
 206             dy10             = _mm_sub_ps(iy1,jy0);
 207             dz10             = _mm_sub_ps(iz1,jz0);
 208             dx20             = _mm_sub_ps(ix2,jx0);
 209             dy20             = _mm_sub_ps(iy2,jy0);
 210             dz20             = _mm_sub_ps(iz2,jz0);
 211             dx30             = _mm_sub_ps(ix3,jx0);
 212             dy30             = _mm_sub_ps(iy3,jy0);
 213             dz30             = _mm_sub_ps(iz3,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 217             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 218             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 219             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 220
 221             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 222             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 223             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 224
 225             rinvsq00         = gmx_mm_inv_ps(rsq00);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 229                                                               charge+jnrC+0,charge+jnrD+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 241                                          vdwparam+vdwioffset0+vdwjidx0B,
 242                                          vdwparam+vdwioffset0+vdwjidx0C,
 243                                          vdwparam+vdwioffset0+vdwjidx0D,
 244                                          &c6_00,&c12_00);
 245
 246             /* LENNARD-JONES DISPERSION/REPULSION */
 247
 248             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 249             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 250             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 251             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 252             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 253
 254             /* Update potential sum for this i atom from the interaction with this j atom. */
 255             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 256
 257             fscal            = fvdw;
 258
 259             /* Calculate temporary vectorial force */
 260             tx               = _mm_mul_ps(fscal,dx00);
 261             ty               = _mm_mul_ps(fscal,dy00);
 262             tz               = _mm_mul_ps(fscal,dz00);
 263
 264             /* Update vectorial force */
 265             fix0             = _mm_add_ps(fix0,tx);
 266             fiy0             = _mm_add_ps(fiy0,ty);
 267             fiz0             = _mm_add_ps(fiz0,tz);
 268
 269             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 270                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 271                                                    tx,ty,tz);
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             r10              = _mm_mul_ps(rsq10,rinv10);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq10             = _mm_mul_ps(iq1,jq0);
 281
 282             /* Calculate table index by multiplying r with table scale and truncate to integer */
 283             rt               = _mm_mul_ps(r10,vftabscale);
 284             vfitab           = _mm_cvttps_epi32(rt);
 285             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 286             vfitab           = _mm_slli_epi32(vfitab,2);
 287
 288             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 289             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 290             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 291             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 292             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 293             _MM_TRANSPOSE4_PS(Y,F,G,H);
 294             Heps             = _mm_mul_ps(vfeps,H);
 295             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 296             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 297             velec            = _mm_mul_ps(qq10,VV);
 298             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 299             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velecsum         = _mm_add_ps(velecsum,velec);
 303
 304             fscal            = felec;
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm_mul_ps(fscal,dx10);
 308             ty               = _mm_mul_ps(fscal,dy10);
 309             tz               = _mm_mul_ps(fscal,dz10);
 310
 311             /* Update vectorial force */
 312             fix1             = _mm_add_ps(fix1,tx);
 313             fiy1             = _mm_add_ps(fiy1,ty);
 314             fiz1             = _mm_add_ps(fiz1,tz);
 315
 316             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 317                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 318                                                    tx,ty,tz);
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             r20              = _mm_mul_ps(rsq20,rinv20);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq20             = _mm_mul_ps(iq2,jq0);
 328
 329             /* Calculate table index by multiplying r with table scale and truncate to integer */
 330             rt               = _mm_mul_ps(r20,vftabscale);
 331             vfitab           = _mm_cvttps_epi32(rt);
 332             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 333             vfitab           = _mm_slli_epi32(vfitab,2);
 334
 335             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 336             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 337             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 338             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 339             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 340             _MM_TRANSPOSE4_PS(Y,F,G,H);
 341             Heps             = _mm_mul_ps(vfeps,H);
 342             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 343             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 344             velec            = _mm_mul_ps(qq20,VV);
 345             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 346             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm_mul_ps(fscal,dx20);
 355             ty               = _mm_mul_ps(fscal,dy20);
 356             tz               = _mm_mul_ps(fscal,dz20);
 357
 358             /* Update vectorial force */
 359             fix2             = _mm_add_ps(fix2,tx);
 360             fiy2             = _mm_add_ps(fiy2,ty);
 361             fiz2             = _mm_add_ps(fiz2,tz);
 362
 363             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 364                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 365                                                    tx,ty,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             r30              = _mm_mul_ps(rsq30,rinv30);
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq30             = _mm_mul_ps(iq3,jq0);
 375
 376             /* Calculate table index by multiplying r with table scale and truncate to integer */
 377             rt               = _mm_mul_ps(r30,vftabscale);
 378             vfitab           = _mm_cvttps_epi32(rt);
 379             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 380             vfitab           = _mm_slli_epi32(vfitab,2);
 381
 382             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 383             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 384             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 385             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 386             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 387             _MM_TRANSPOSE4_PS(Y,F,G,H);
 388             Heps             = _mm_mul_ps(vfeps,H);
 389             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 390             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 391             velec            = _mm_mul_ps(qq30,VV);
 392             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 393             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velecsum         = _mm_add_ps(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             /* Calculate temporary vectorial force */
 401             tx               = _mm_mul_ps(fscal,dx30);
 402             ty               = _mm_mul_ps(fscal,dy30);
 403             tz               = _mm_mul_ps(fscal,dz30);
 404
 405             /* Update vectorial force */
 406             fix3             = _mm_add_ps(fix3,tx);
 407             fiy3             = _mm_add_ps(fiy3,ty);
 408             fiz3             = _mm_add_ps(fiz3,tz);
 409
 410             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 411                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 412                                                    tx,ty,tz);
 413
 414             /* Inner loop uses 161 flops */
 415         }
 416
 417         if(jidx<j_index_end)
 418         {
 419
 420             /* Get j neighbor index, and coordinate index */
 421             jnrA             = jjnr[jidx];
 422             jnrB             = jjnr[jidx+1];
 423             jnrC             = jjnr[jidx+2];
 424             jnrD             = jjnr[jidx+3];
 425
 426             /* Sign of each element will be negative for non-real atoms.
 427              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 428              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 429              */
 430             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 431             jnrA       = (jnrA>=0) ? jnrA : 0;
 432             jnrB       = (jnrB>=0) ? jnrB : 0;
 433             jnrC       = (jnrC>=0) ? jnrC : 0;
 434             jnrD       = (jnrD>=0) ? jnrD : 0;
 435
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438             j_coord_offsetC  = DIM*jnrC;
 439             j_coord_offsetD  = DIM*jnrD;
 440
 441             /* load j atom coordinates */
 442             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 443                                               x+j_coord_offsetC,x+j_coord_offsetD,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm_sub_ps(ix0,jx0);
 448             dy00             = _mm_sub_ps(iy0,jy0);
 449             dz00             = _mm_sub_ps(iz0,jz0);
 450             dx10             = _mm_sub_ps(ix1,jx0);
 451             dy10             = _mm_sub_ps(iy1,jy0);
 452             dz10             = _mm_sub_ps(iz1,jz0);
 453             dx20             = _mm_sub_ps(ix2,jx0);
 454             dy20             = _mm_sub_ps(iy2,jy0);
 455             dz20             = _mm_sub_ps(iz2,jz0);
 456             dx30             = _mm_sub_ps(ix3,jx0);
 457             dy30             = _mm_sub_ps(iy3,jy0);
 458             dz30             = _mm_sub_ps(iz3,jz0);
 459
 460             /* Calculate squared distance and things based on it */
 461             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 462             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 463             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 464             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 465
 466             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 467             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 468             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 469
 470             rinvsq00         = gmx_mm_inv_ps(rsq00);
 471
 472             /* Load parameters for j particles */
 473             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 474                                                               charge+jnrC+0,charge+jnrD+0);
 475             vdwjidx0A        = 2*vdwtype[jnrA+0];
 476             vdwjidx0B        = 2*vdwtype[jnrB+0];
 477             vdwjidx0C        = 2*vdwtype[jnrC+0];
 478             vdwjidx0D        = 2*vdwtype[jnrD+0];
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 486                                          vdwparam+vdwioffset0+vdwjidx0B,
 487                                          vdwparam+vdwioffset0+vdwjidx0C,
 488                                          vdwparam+vdwioffset0+vdwjidx0D,
 489                                          &c6_00,&c12_00);
 490
 491             /* LENNARD-JONES DISPERSION/REPULSION */
 492
 493             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 494             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 495             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 496             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 497             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 498
 499             /* Update potential sum for this i atom from the interaction with this j atom. */
 500             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 501             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 502
 503             fscal            = fvdw;
 504
 505             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 506
 507             /* Calculate temporary vectorial force */
 508             tx               = _mm_mul_ps(fscal,dx00);
 509             ty               = _mm_mul_ps(fscal,dy00);
 510             tz               = _mm_mul_ps(fscal,dz00);
 511
 512             /* Update vectorial force */
 513             fix0             = _mm_add_ps(fix0,tx);
 514             fiy0             = _mm_add_ps(fiy0,ty);
 515             fiz0             = _mm_add_ps(fiz0,tz);
 516
 517             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 518                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 519                                                    tx,ty,tz);
 520
 521             /**************************
 522              * CALCULATE INTERACTIONS *
 523              **************************/
 524
 525             r10              = _mm_mul_ps(rsq10,rinv10);
 526             r10              = _mm_andnot_ps(dummy_mask,r10);
 527
 528             /* Compute parameters for interactions between i and j atoms */
 529             qq10             = _mm_mul_ps(iq1,jq0);
 530
 531             /* Calculate table index by multiplying r with table scale and truncate to integer */
 532             rt               = _mm_mul_ps(r10,vftabscale);
 533             vfitab           = _mm_cvttps_epi32(rt);
 534             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 535             vfitab           = _mm_slli_epi32(vfitab,2);
 536
 537             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 538             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 539             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 540             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 541             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 542             _MM_TRANSPOSE4_PS(Y,F,G,H);
 543             Heps             = _mm_mul_ps(vfeps,H);
 544             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 545             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 546             velec            = _mm_mul_ps(qq10,VV);
 547             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 548             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _mm_andnot_ps(dummy_mask,velec);
 552             velecsum         = _mm_add_ps(velecsum,velec);
 553
 554             fscal            = felec;
 555
 556             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 557
 558             /* Calculate temporary vectorial force */
 559             tx               = _mm_mul_ps(fscal,dx10);
 560             ty               = _mm_mul_ps(fscal,dy10);
 561             tz               = _mm_mul_ps(fscal,dz10);
 562
 563             /* Update vectorial force */
 564             fix1             = _mm_add_ps(fix1,tx);
 565             fiy1             = _mm_add_ps(fiy1,ty);
 566             fiz1             = _mm_add_ps(fiz1,tz);
 567
 568             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 569                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 570                                                    tx,ty,tz);
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             r20              = _mm_mul_ps(rsq20,rinv20);
 577             r20              = _mm_andnot_ps(dummy_mask,r20);
 578
 579             /* Compute parameters for interactions between i and j atoms */
 580             qq20             = _mm_mul_ps(iq2,jq0);
 581
 582             /* Calculate table index by multiplying r with table scale and truncate to integer */
 583             rt               = _mm_mul_ps(r20,vftabscale);
 584             vfitab           = _mm_cvttps_epi32(rt);
 585             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 586             vfitab           = _mm_slli_epi32(vfitab,2);
 587
 588             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 589             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 590             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 591             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 592             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 593             _MM_TRANSPOSE4_PS(Y,F,G,H);
 594             Heps             = _mm_mul_ps(vfeps,H);
 595             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 596             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 597             velec            = _mm_mul_ps(qq20,VV);
 598             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 599             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 600
 601             /* Update potential sum for this i atom from the interaction with this j atom. */
 602             velec            = _mm_andnot_ps(dummy_mask,velec);
 603             velecsum         = _mm_add_ps(velecsum,velec);
 604
 605             fscal            = felec;
 606
 607             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm_mul_ps(fscal,dx20);
 611             ty               = _mm_mul_ps(fscal,dy20);
 612             tz               = _mm_mul_ps(fscal,dz20);
 613
 614             /* Update vectorial force */
 615             fix2             = _mm_add_ps(fix2,tx);
 616             fiy2             = _mm_add_ps(fiy2,ty);
 617             fiz2             = _mm_add_ps(fiz2,tz);
 618
 619             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 620                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 621                                                    tx,ty,tz);
 622
 623             /**************************
 624              * CALCULATE INTERACTIONS *
 625              **************************/
 626
 627             r30              = _mm_mul_ps(rsq30,rinv30);
 628             r30              = _mm_andnot_ps(dummy_mask,r30);
 629
 630             /* Compute parameters for interactions between i and j atoms */
 631             qq30             = _mm_mul_ps(iq3,jq0);
 632
 633             /* Calculate table index by multiplying r with table scale and truncate to integer */
 634             rt               = _mm_mul_ps(r30,vftabscale);
 635             vfitab           = _mm_cvttps_epi32(rt);
 636             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 637             vfitab           = _mm_slli_epi32(vfitab,2);
 638
 639             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 640             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 641             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 642             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 643             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 644             _MM_TRANSPOSE4_PS(Y,F,G,H);
 645             Heps             = _mm_mul_ps(vfeps,H);
 646             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 647             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 648             velec            = _mm_mul_ps(qq30,VV);
 649             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 650             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 651
 652             /* Update potential sum for this i atom from the interaction with this j atom. */
 653             velec            = _mm_andnot_ps(dummy_mask,velec);
 654             velecsum         = _mm_add_ps(velecsum,velec);
 655
 656             fscal            = felec;
 657
 658             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 659
 660             /* Calculate temporary vectorial force */
 661             tx               = _mm_mul_ps(fscal,dx30);
 662             ty               = _mm_mul_ps(fscal,dy30);
 663             tz               = _mm_mul_ps(fscal,dz30);
 664
 665             /* Update vectorial force */
 666             fix3             = _mm_add_ps(fix3,tx);
 667             fiy3             = _mm_add_ps(fiy3,ty);
 668             fiz3             = _mm_add_ps(fiz3,tz);
 669
 670             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 671                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 672                                                    tx,ty,tz);
 673
 674             /* Inner loop uses 164 flops */
 675         }
 676
 677         /* End of innermost loop */
 678
 679         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 680                                               f+i_coord_offset,fshift+i_shift_offset);
 681
 682         ggid                        = gid[iidx];
 683         /* Update potential energies */
 684         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 685         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 686
 687         /* Increment number of inner iterations */
 688         inneriter                  += j_index_end - j_index_start;
 689
 690         /* Outer loop uses 38 flops */
 691     }
 692
 693     /* Increment number of outer iterations */
 694     outeriter        += nri;
 695
 696     /* Update outer/inner flops */
 697
 698     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*38 + inneriter*164);
 699 }
 700 /*
 701  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_single
 702  * Electrostatics interaction: CubicSplineTable
 703  * VdW interaction:            LennardJones
 704  * Geometry:                   Water4-Particle
 705  * Calculate force/pot:        Force
 706  */
 707 void
 708 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_single
 709                     (t_nblist * gmx_restrict                nlist,
 710                      rvec * gmx_restrict                    xx,
 711                      rvec * gmx_restrict                    ff,
 712                      t_forcerec * gmx_restrict              fr,
 713                      t_mdatoms * gmx_restrict               mdatoms,
 714                      nb_kernel_data_t * gmx_restrict        kernel_data,
 715                      t_nrnb * gmx_restrict                  nrnb)
 716 {
 717     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 718      * just 0 for non-waters.
 719      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 720      * jnr indices corresponding to data put in the four positions in the SIMD register.
 721      */
 722     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 723     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 724     int              jnrA,jnrB,jnrC,jnrD;
 725     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 726     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 727     real             shX,shY,shZ,rcutoff_scalar;
 728     real             *shiftvec,*fshift,*x,*f;
 729     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 730     int              vdwioffset0;
 731     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 732     int              vdwioffset1;
 733     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 734     int              vdwioffset2;
 735     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 736     int              vdwioffset3;
 737     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 738     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 739     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 740     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 741     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 742     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 743     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 744     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 745     real             *charge;
 746     int              nvdwtype;
 747     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 748     int              *vdwtype;
 749     real             *vdwparam;
 750     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 751     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 752     __m128i          vfitab;
 753     __m128i          ifour       = _mm_set1_epi32(4);
 754     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 755     real             *vftab;
 756     __m128           dummy_mask,cutoff_mask;
 757     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 758     __m128           one     = _mm_set1_ps(1.0);
 759     __m128           two     = _mm_set1_ps(2.0);
 760     x                = xx[0];
 761     f                = ff[0];
 762
 763     nri              = nlist->nri;
 764     iinr             = nlist->iinr;
 765     jindex           = nlist->jindex;
 766     jjnr             = nlist->jjnr;
 767     shiftidx         = nlist->shift;
 768     gid              = nlist->gid;
 769     shiftvec         = fr->shift_vec[0];
 770     fshift           = fr->fshift[0];
 771     facel            = _mm_set1_ps(fr->epsfac);
 772     charge           = mdatoms->chargeA;
 773     nvdwtype         = fr->ntype;
 774     vdwparam         = fr->nbfp;
 775     vdwtype          = mdatoms->typeA;
 776
 777     vftab            = kernel_data->table_elec->data;
 778     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 779
 780     /* Setup water-specific parameters */
 781     inr              = nlist->iinr[0];
 782     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 783     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 784     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 785     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 786
 787     /* Avoid stupid compiler warnings */
 788     jnrA = jnrB = jnrC = jnrD = 0;
 789     j_coord_offsetA = 0;
 790     j_coord_offsetB = 0;
 791     j_coord_offsetC = 0;
 792     j_coord_offsetD = 0;
 793
 794     outeriter        = 0;
 795     inneriter        = 0;
 796
 797     /* Start outer loop over neighborlists */
 798     for(iidx=0; iidx<nri; iidx++)
 799     {
 800         /* Load shift vector for this list */
 801         i_shift_offset   = DIM*shiftidx[iidx];
 802         shX              = shiftvec[i_shift_offset+XX];
 803         shY              = shiftvec[i_shift_offset+YY];
 804         shZ              = shiftvec[i_shift_offset+ZZ];
 805
 806         /* Load limits for loop over neighbors */
 807         j_index_start    = jindex[iidx];
 808         j_index_end      = jindex[iidx+1];
 809
 810         /* Get outer coordinate index */
 811         inr              = iinr[iidx];
 812         i_coord_offset   = DIM*inr;
 813
 814         /* Load i particle coords and add shift vector */
 815         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 816         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 817         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 818         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 819         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 820         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 821         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 822         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 823         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 824         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 825         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 826         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 827
 828         fix0             = _mm_setzero_ps();
 829         fiy0             = _mm_setzero_ps();
 830         fiz0             = _mm_setzero_ps();
 831         fix1             = _mm_setzero_ps();
 832         fiy1             = _mm_setzero_ps();
 833         fiz1             = _mm_setzero_ps();
 834         fix2             = _mm_setzero_ps();
 835         fiy2             = _mm_setzero_ps();
 836         fiz2             = _mm_setzero_ps();
 837         fix3             = _mm_setzero_ps();
 838         fiy3             = _mm_setzero_ps();
 839         fiz3             = _mm_setzero_ps();
 840
 841         /* Start inner kernel loop */
 842         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 843         {
 844
 845             /* Get j neighbor index, and coordinate index */
 846             jnrA             = jjnr[jidx];
 847             jnrB             = jjnr[jidx+1];
 848             jnrC             = jjnr[jidx+2];
 849             jnrD             = jjnr[jidx+3];
 850
 851             j_coord_offsetA  = DIM*jnrA;
 852             j_coord_offsetB  = DIM*jnrB;
 853             j_coord_offsetC  = DIM*jnrC;
 854             j_coord_offsetD  = DIM*jnrD;
 855
 856             /* load j atom coordinates */
 857             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 858                                               x+j_coord_offsetC,x+j_coord_offsetD,
 859                                               &jx0,&jy0,&jz0);
 860
 861             /* Calculate displacement vector */
 862             dx00             = _mm_sub_ps(ix0,jx0);
 863             dy00             = _mm_sub_ps(iy0,jy0);
 864             dz00             = _mm_sub_ps(iz0,jz0);
 865             dx10             = _mm_sub_ps(ix1,jx0);
 866             dy10             = _mm_sub_ps(iy1,jy0);
 867             dz10             = _mm_sub_ps(iz1,jz0);
 868             dx20             = _mm_sub_ps(ix2,jx0);
 869             dy20             = _mm_sub_ps(iy2,jy0);
 870             dz20             = _mm_sub_ps(iz2,jz0);
 871             dx30             = _mm_sub_ps(ix3,jx0);
 872             dy30             = _mm_sub_ps(iy3,jy0);
 873             dz30             = _mm_sub_ps(iz3,jz0);
 874
 875             /* Calculate squared distance and things based on it */
 876             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 877             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 878             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 879             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 880
 881             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 882             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 883             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 884
 885             rinvsq00         = gmx_mm_inv_ps(rsq00);
 886
 887             /* Load parameters for j particles */
 888             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 889                                                               charge+jnrC+0,charge+jnrD+0);
 890             vdwjidx0A        = 2*vdwtype[jnrA+0];
 891             vdwjidx0B        = 2*vdwtype[jnrB+0];
 892             vdwjidx0C        = 2*vdwtype[jnrC+0];
 893             vdwjidx0D        = 2*vdwtype[jnrD+0];
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             /* Compute parameters for interactions between i and j atoms */
 900             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 901                                          vdwparam+vdwioffset0+vdwjidx0B,
 902                                          vdwparam+vdwioffset0+vdwjidx0C,
 903                                          vdwparam+vdwioffset0+vdwjidx0D,
 904                                          &c6_00,&c12_00);
 905
 906             /* LENNARD-JONES DISPERSION/REPULSION */
 907
 908             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 909             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 910
 911             fscal            = fvdw;
 912
 913             /* Calculate temporary vectorial force */
 914             tx               = _mm_mul_ps(fscal,dx00);
 915             ty               = _mm_mul_ps(fscal,dy00);
 916             tz               = _mm_mul_ps(fscal,dz00);
 917
 918             /* Update vectorial force */
 919             fix0             = _mm_add_ps(fix0,tx);
 920             fiy0             = _mm_add_ps(fiy0,ty);
 921             fiz0             = _mm_add_ps(fiz0,tz);
 922
 923             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 924                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 925                                                    tx,ty,tz);
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             r10              = _mm_mul_ps(rsq10,rinv10);
 932
 933             /* Compute parameters for interactions between i and j atoms */
 934             qq10             = _mm_mul_ps(iq1,jq0);
 935
 936             /* Calculate table index by multiplying r with table scale and truncate to integer */
 937             rt               = _mm_mul_ps(r10,vftabscale);
 938             vfitab           = _mm_cvttps_epi32(rt);
 939             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 940             vfitab           = _mm_slli_epi32(vfitab,2);
 941
 942             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 943             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 944             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 945             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 946             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 947             _MM_TRANSPOSE4_PS(Y,F,G,H);
 948             Heps             = _mm_mul_ps(vfeps,H);
 949             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 950             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 951             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 952
 953             fscal            = felec;
 954
 955             /* Calculate temporary vectorial force */
 956             tx               = _mm_mul_ps(fscal,dx10);
 957             ty               = _mm_mul_ps(fscal,dy10);
 958             tz               = _mm_mul_ps(fscal,dz10);
 959
 960             /* Update vectorial force */
 961             fix1             = _mm_add_ps(fix1,tx);
 962             fiy1             = _mm_add_ps(fiy1,ty);
 963             fiz1             = _mm_add_ps(fiz1,tz);
 964
 965             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 966                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 967                                                    tx,ty,tz);
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             r20              = _mm_mul_ps(rsq20,rinv20);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq20             = _mm_mul_ps(iq2,jq0);
 977
 978             /* Calculate table index by multiplying r with table scale and truncate to integer */
 979             rt               = _mm_mul_ps(r20,vftabscale);
 980             vfitab           = _mm_cvttps_epi32(rt);
 981             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 982             vfitab           = _mm_slli_epi32(vfitab,2);
 983
 984             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 985             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 986             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 987             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 988             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 989             _MM_TRANSPOSE4_PS(Y,F,G,H);
 990             Heps             = _mm_mul_ps(vfeps,H);
 991             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 992             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 993             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 994
 995             fscal            = felec;
 996
 997             /* Calculate temporary vectorial force */
 998             tx               = _mm_mul_ps(fscal,dx20);
 999             ty               = _mm_mul_ps(fscal,dy20);
1000             tz               = _mm_mul_ps(fscal,dz20);
1001
1002             /* Update vectorial force */
1003             fix2             = _mm_add_ps(fix2,tx);
1004             fiy2             = _mm_add_ps(fiy2,ty);
1005             fiz2             = _mm_add_ps(fiz2,tz);
1006
1007             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1008                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1009                                                    tx,ty,tz);
1010
1011             /**************************
1012              * CALCULATE INTERACTIONS *
1013              **************************/
1014
1015             r30              = _mm_mul_ps(rsq30,rinv30);
1016
1017             /* Compute parameters for interactions between i and j atoms */
1018             qq30             = _mm_mul_ps(iq3,jq0);
1019
1020             /* Calculate table index by multiplying r with table scale and truncate to integer */
1021             rt               = _mm_mul_ps(r30,vftabscale);
1022             vfitab           = _mm_cvttps_epi32(rt);
1023             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1024             vfitab           = _mm_slli_epi32(vfitab,2);
1025
1026             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1027             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1028             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1029             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1030             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1031             _MM_TRANSPOSE4_PS(Y,F,G,H);
1032             Heps             = _mm_mul_ps(vfeps,H);
1033             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1034             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1035             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1036
1037             fscal            = felec;
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm_mul_ps(fscal,dx30);
1041             ty               = _mm_mul_ps(fscal,dy30);
1042             tz               = _mm_mul_ps(fscal,dz30);
1043
1044             /* Update vectorial force */
1045             fix3             = _mm_add_ps(fix3,tx);
1046             fiy3             = _mm_add_ps(fiy3,ty);
1047             fiz3             = _mm_add_ps(fiz3,tz);
1048
1049             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1050                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1051                                                    tx,ty,tz);
1052
1053             /* Inner loop uses 144 flops */
1054         }
1055
1056         if(jidx<j_index_end)
1057         {
1058
1059             /* Get j neighbor index, and coordinate index */
1060             jnrA             = jjnr[jidx];
1061             jnrB             = jjnr[jidx+1];
1062             jnrC             = jjnr[jidx+2];
1063             jnrD             = jjnr[jidx+3];
1064
1065             /* Sign of each element will be negative for non-real atoms.
1066              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1067              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1068              */
1069             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1070             jnrA       = (jnrA>=0) ? jnrA : 0;
1071             jnrB       = (jnrB>=0) ? jnrB : 0;
1072             jnrC       = (jnrC>=0) ? jnrC : 0;
1073             jnrD       = (jnrD>=0) ? jnrD : 0;
1074
1075             j_coord_offsetA  = DIM*jnrA;
1076             j_coord_offsetB  = DIM*jnrB;
1077             j_coord_offsetC  = DIM*jnrC;
1078             j_coord_offsetD  = DIM*jnrD;
1079
1080             /* load j atom coordinates */
1081             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1082                                               x+j_coord_offsetC,x+j_coord_offsetD,
1083                                               &jx0,&jy0,&jz0);
1084
1085             /* Calculate displacement vector */
1086             dx00             = _mm_sub_ps(ix0,jx0);
1087             dy00             = _mm_sub_ps(iy0,jy0);
1088             dz00             = _mm_sub_ps(iz0,jz0);
1089             dx10             = _mm_sub_ps(ix1,jx0);
1090             dy10             = _mm_sub_ps(iy1,jy0);
1091             dz10             = _mm_sub_ps(iz1,jz0);
1092             dx20             = _mm_sub_ps(ix2,jx0);
1093             dy20             = _mm_sub_ps(iy2,jy0);
1094             dz20             = _mm_sub_ps(iz2,jz0);
1095             dx30             = _mm_sub_ps(ix3,jx0);
1096             dy30             = _mm_sub_ps(iy3,jy0);
1097             dz30             = _mm_sub_ps(iz3,jz0);
1098
1099             /* Calculate squared distance and things based on it */
1100             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1101             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1102             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1103             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1104
1105             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1106             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1107             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1108
1109             rinvsq00         = gmx_mm_inv_ps(rsq00);
1110
1111             /* Load parameters for j particles */
1112             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1113                                                               charge+jnrC+0,charge+jnrD+0);
1114             vdwjidx0A        = 2*vdwtype[jnrA+0];
1115             vdwjidx0B        = 2*vdwtype[jnrB+0];
1116             vdwjidx0C        = 2*vdwtype[jnrC+0];
1117             vdwjidx0D        = 2*vdwtype[jnrD+0];
1118
1119             /**************************
1120              * CALCULATE INTERACTIONS *
1121              **************************/
1122
1123             /* Compute parameters for interactions between i and j atoms */
1124             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1125                                          vdwparam+vdwioffset0+vdwjidx0B,
1126                                          vdwparam+vdwioffset0+vdwjidx0C,
1127                                          vdwparam+vdwioffset0+vdwjidx0D,
1128                                          &c6_00,&c12_00);
1129
1130             /* LENNARD-JONES DISPERSION/REPULSION */
1131
1132             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1133             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1134
1135             fscal            = fvdw;
1136
1137             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm_mul_ps(fscal,dx00);
1141             ty               = _mm_mul_ps(fscal,dy00);
1142             tz               = _mm_mul_ps(fscal,dz00);
1143
1144             /* Update vectorial force */
1145             fix0             = _mm_add_ps(fix0,tx);
1146             fiy0             = _mm_add_ps(fiy0,ty);
1147             fiz0             = _mm_add_ps(fiz0,tz);
1148
1149             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1150                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1151                                                    tx,ty,tz);
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             r10              = _mm_mul_ps(rsq10,rinv10);
1158             r10              = _mm_andnot_ps(dummy_mask,r10);
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             qq10             = _mm_mul_ps(iq1,jq0);
1162
1163             /* Calculate table index by multiplying r with table scale and truncate to integer */
1164             rt               = _mm_mul_ps(r10,vftabscale);
1165             vfitab           = _mm_cvttps_epi32(rt);
1166             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1167             vfitab           = _mm_slli_epi32(vfitab,2);
1168
1169             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1170             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1171             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1172             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1173             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1174             _MM_TRANSPOSE4_PS(Y,F,G,H);
1175             Heps             = _mm_mul_ps(vfeps,H);
1176             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1177             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1178             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1179
1180             fscal            = felec;
1181
1182             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1183
1184             /* Calculate temporary vectorial force */
1185             tx               = _mm_mul_ps(fscal,dx10);
1186             ty               = _mm_mul_ps(fscal,dy10);
1187             tz               = _mm_mul_ps(fscal,dz10);
1188
1189             /* Update vectorial force */
1190             fix1             = _mm_add_ps(fix1,tx);
1191             fiy1             = _mm_add_ps(fiy1,ty);
1192             fiz1             = _mm_add_ps(fiz1,tz);
1193
1194             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1195                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1196                                                    tx,ty,tz);
1197
1198             /**************************
1199              * CALCULATE INTERACTIONS *
1200              **************************/
1201
1202             r20              = _mm_mul_ps(rsq20,rinv20);
1203             r20              = _mm_andnot_ps(dummy_mask,r20);
1204
1205             /* Compute parameters for interactions between i and j atoms */
1206             qq20             = _mm_mul_ps(iq2,jq0);
1207
1208             /* Calculate table index by multiplying r with table scale and truncate to integer */
1209             rt               = _mm_mul_ps(r20,vftabscale);
1210             vfitab           = _mm_cvttps_epi32(rt);
1211             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1212             vfitab           = _mm_slli_epi32(vfitab,2);
1213
1214             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1215             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1216             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1217             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1218             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1219             _MM_TRANSPOSE4_PS(Y,F,G,H);
1220             Heps             = _mm_mul_ps(vfeps,H);
1221             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1222             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1223             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1224
1225             fscal            = felec;
1226
1227             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1228
1229             /* Calculate temporary vectorial force */
1230             tx               = _mm_mul_ps(fscal,dx20);
1231             ty               = _mm_mul_ps(fscal,dy20);
1232             tz               = _mm_mul_ps(fscal,dz20);
1233
1234             /* Update vectorial force */
1235             fix2             = _mm_add_ps(fix2,tx);
1236             fiy2             = _mm_add_ps(fiy2,ty);
1237             fiz2             = _mm_add_ps(fiz2,tz);
1238
1239             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1240                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1241                                                    tx,ty,tz);
1242
1243             /**************************
1244              * CALCULATE INTERACTIONS *
1245              **************************/
1246
1247             r30              = _mm_mul_ps(rsq30,rinv30);
1248             r30              = _mm_andnot_ps(dummy_mask,r30);
1249
1250             /* Compute parameters for interactions between i and j atoms */
1251             qq30             = _mm_mul_ps(iq3,jq0);
1252
1253             /* Calculate table index by multiplying r with table scale and truncate to integer */
1254             rt               = _mm_mul_ps(r30,vftabscale);
1255             vfitab           = _mm_cvttps_epi32(rt);
1256             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1257             vfitab           = _mm_slli_epi32(vfitab,2);
1258
1259             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1260             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1261             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1262             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1263             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1264             _MM_TRANSPOSE4_PS(Y,F,G,H);
1265             Heps             = _mm_mul_ps(vfeps,H);
1266             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1267             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1268             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1269
1270             fscal            = felec;
1271
1272             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1273
1274             /* Calculate temporary vectorial force */
1275             tx               = _mm_mul_ps(fscal,dx30);
1276             ty               = _mm_mul_ps(fscal,dy30);
1277             tz               = _mm_mul_ps(fscal,dz30);
1278
1279             /* Update vectorial force */
1280             fix3             = _mm_add_ps(fix3,tx);
1281             fiy3             = _mm_add_ps(fiy3,ty);
1282             fiz3             = _mm_add_ps(fiz3,tz);
1283
1284             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1285                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1286                                                    tx,ty,tz);
1287
1288             /* Inner loop uses 147 flops */
1289         }
1290
1291         /* End of innermost loop */
1292
1293         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1294                                               f+i_coord_offset,fshift+i_shift_offset);
1295
1296         /* Increment number of inner iterations */
1297         inneriter                  += j_index_end - j_index_start;
1298
1299         /* Outer loop uses 36 flops */
1300     }
1301
1302     /* Increment number of outer iterations */
1303     outeriter        += nri;
1304
1305     /* Update outer/inner flops */
1306
1307     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*36 + inneriter*147);
1308 }