src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_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_ElecEw_VdwCSTab_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  84     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  85     __m128i          vfitab;
  86     __m128i          ifour       = _mm_set1_epi32(4);
  87     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  88     real             *vftab;
  89     __m128i          ewitab;
  90     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  91     real             *ewtab;
  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_vdw->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 115
 116     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 117     ewtab            = fr->ic->tabq_coul_FDV0;
 118     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 119     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 124     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 125     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 126     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143         shX              = shiftvec[i_shift_offset+XX];
 144         shY              = shiftvec[i_shift_offset+YY];
 145         shZ              = shiftvec[i_shift_offset+ZZ];
 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         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 157         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 158         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 159         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 160         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 161         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 162         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 163         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 164         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 165
 166         fix0             = _mm_setzero_ps();
 167         fiy0             = _mm_setzero_ps();
 168         fiz0             = _mm_setzero_ps();
 169         fix1             = _mm_setzero_ps();
 170         fiy1             = _mm_setzero_ps();
 171         fiz1             = _mm_setzero_ps();
 172         fix2             = _mm_setzero_ps();
 173         fiy2             = _mm_setzero_ps();
 174         fiz2             = _mm_setzero_ps();
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_ps();
 178         vvdwsum          = _mm_setzero_ps();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             jnrC             = jjnr[jidx+2];
 188             jnrD             = jjnr[jidx+3];
 189
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192             j_coord_offsetC  = DIM*jnrC;
 193             j_coord_offsetD  = DIM*jnrD;
 194
 195             /* load j atom coordinates */
 196             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                               x+j_coord_offsetC,x+j_coord_offsetD,
 198                                               &jx0,&jy0,&jz0);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm_sub_ps(ix0,jx0);
 202             dy00             = _mm_sub_ps(iy0,jy0);
 203             dz00             = _mm_sub_ps(iz0,jz0);
 204             dx10             = _mm_sub_ps(ix1,jx0);
 205             dy10             = _mm_sub_ps(iy1,jy0);
 206             dz10             = _mm_sub_ps(iz1,jz0);
 207             dx20             = _mm_sub_ps(ix2,jx0);
 208             dy20             = _mm_sub_ps(iy2,jy0);
 209             dz20             = _mm_sub_ps(iz2,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 213             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 215
 216             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 217             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 218             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 219
 220             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 221             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 226                                                               charge+jnrC+0,charge+jnrD+0);
 227             vdwjidx0A        = 2*vdwtype[jnrA+0];
 228             vdwjidx0B        = 2*vdwtype[jnrB+0];
 229             vdwjidx0C        = 2*vdwtype[jnrC+0];
 230             vdwjidx0D        = 2*vdwtype[jnrD+0];
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             r00              = _mm_mul_ps(rsq00,rinv00);
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq00             = _mm_mul_ps(iq0,jq0);
 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             /* Calculate table index by multiplying r with table scale and truncate to integer */
 247             rt               = _mm_mul_ps(r00,vftabscale);
 248             vfitab           = _mm_cvttps_epi32(rt);
 249             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 250             vfitab           = _mm_slli_epi32(vfitab,3);
 251
 252             /* EWALD ELECTROSTATICS */
 253
 254             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 255             ewrt             = _mm_mul_ps(r00,ewtabscale);
 256             ewitab           = _mm_cvttps_epi32(ewrt);
 257             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 258             ewitab           = _mm_slli_epi32(ewitab,2);
 259             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 260             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 261             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 262             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 263             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 264             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 265             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 266             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 267             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 268
 269             /* CUBIC SPLINE TABLE DISPERSION */
 270             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 271             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 272             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 273             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 274             _MM_TRANSPOSE4_PS(Y,F,G,H);
 275             Heps             = _mm_mul_ps(vfeps,H);
 276             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 277             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 278             vvdw6            = _mm_mul_ps(c6_00,VV);
 279             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 280             fvdw6            = _mm_mul_ps(c6_00,FF);
 281
 282             /* CUBIC SPLINE TABLE REPULSION */
 283             vfitab           = _mm_add_epi32(vfitab,ifour);
 284             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 285             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 286             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 287             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 288             _MM_TRANSPOSE4_PS(Y,F,G,H);
 289             Heps             = _mm_mul_ps(vfeps,H);
 290             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 291             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 292             vvdw12           = _mm_mul_ps(c12_00,VV);
 293             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 294             fvdw12           = _mm_mul_ps(c12_00,FF);
 295             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 296             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velecsum         = _mm_add_ps(velecsum,velec);
 300             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 301
 302             fscal            = _mm_add_ps(felec,fvdw);
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm_mul_ps(fscal,dx00);
 306             ty               = _mm_mul_ps(fscal,dy00);
 307             tz               = _mm_mul_ps(fscal,dz00);
 308
 309             /* Update vectorial force */
 310             fix0             = _mm_add_ps(fix0,tx);
 311             fiy0             = _mm_add_ps(fiy0,ty);
 312             fiz0             = _mm_add_ps(fiz0,tz);
 313
 314             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 315                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 316                                                    tx,ty,tz);
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             r10              = _mm_mul_ps(rsq10,rinv10);
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq10             = _mm_mul_ps(iq1,jq0);
 326
 327             /* EWALD ELECTROSTATICS */
 328
 329             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 330             ewrt             = _mm_mul_ps(r10,ewtabscale);
 331             ewitab           = _mm_cvttps_epi32(ewrt);
 332             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 333             ewitab           = _mm_slli_epi32(ewitab,2);
 334             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 335             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 336             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 337             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 338             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 339             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 340             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 341             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 342             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm_add_ps(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm_mul_ps(fscal,dx10);
 351             ty               = _mm_mul_ps(fscal,dy10);
 352             tz               = _mm_mul_ps(fscal,dz10);
 353
 354             /* Update vectorial force */
 355             fix1             = _mm_add_ps(fix1,tx);
 356             fiy1             = _mm_add_ps(fiy1,ty);
 357             fiz1             = _mm_add_ps(fiz1,tz);
 358
 359             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 360                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 361                                                    tx,ty,tz);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             r20              = _mm_mul_ps(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm_mul_ps(iq2,jq0);
 371
 372             /* EWALD ELECTROSTATICS */
 373
 374             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 375             ewrt             = _mm_mul_ps(r20,ewtabscale);
 376             ewitab           = _mm_cvttps_epi32(ewrt);
 377             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 378             ewitab           = _mm_slli_epi32(ewitab,2);
 379             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 380             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 381             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 382             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 383             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 384             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 385             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 386             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 387             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velecsum         = _mm_add_ps(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm_mul_ps(fscal,dx20);
 396             ty               = _mm_mul_ps(fscal,dy20);
 397             tz               = _mm_mul_ps(fscal,dz20);
 398
 399             /* Update vectorial force */
 400             fix2             = _mm_add_ps(fix2,tx);
 401             fiy2             = _mm_add_ps(fiy2,ty);
 402             fiz2             = _mm_add_ps(fiz2,tz);
 403
 404             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 405                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 406                                                    tx,ty,tz);
 407
 408             /* Inner loop uses 157 flops */
 409         }
 410
 411         if(jidx<j_index_end)
 412         {
 413
 414             /* Get j neighbor index, and coordinate index */
 415             jnrA             = jjnr[jidx];
 416             jnrB             = jjnr[jidx+1];
 417             jnrC             = jjnr[jidx+2];
 418             jnrD             = jjnr[jidx+3];
 419
 420             /* Sign of each element will be negative for non-real atoms.
 421              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 422              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 423              */
 424             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 425             jnrA       = (jnrA>=0) ? jnrA : 0;
 426             jnrB       = (jnrB>=0) ? jnrB : 0;
 427             jnrC       = (jnrC>=0) ? jnrC : 0;
 428             jnrD       = (jnrD>=0) ? jnrD : 0;
 429
 430             j_coord_offsetA  = DIM*jnrA;
 431             j_coord_offsetB  = DIM*jnrB;
 432             j_coord_offsetC  = DIM*jnrC;
 433             j_coord_offsetD  = DIM*jnrD;
 434
 435             /* load j atom coordinates */
 436             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 437                                               x+j_coord_offsetC,x+j_coord_offsetD,
 438                                               &jx0,&jy0,&jz0);
 439
 440             /* Calculate displacement vector */
 441             dx00             = _mm_sub_ps(ix0,jx0);
 442             dy00             = _mm_sub_ps(iy0,jy0);
 443             dz00             = _mm_sub_ps(iz0,jz0);
 444             dx10             = _mm_sub_ps(ix1,jx0);
 445             dy10             = _mm_sub_ps(iy1,jy0);
 446             dz10             = _mm_sub_ps(iz1,jz0);
 447             dx20             = _mm_sub_ps(ix2,jx0);
 448             dy20             = _mm_sub_ps(iy2,jy0);
 449             dz20             = _mm_sub_ps(iz2,jz0);
 450
 451             /* Calculate squared distance and things based on it */
 452             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 453             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 454             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 455
 456             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 457             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 458             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 459
 460             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 461             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 462             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 463
 464             /* Load parameters for j particles */
 465             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 466                                                               charge+jnrC+0,charge+jnrD+0);
 467             vdwjidx0A        = 2*vdwtype[jnrA+0];
 468             vdwjidx0B        = 2*vdwtype[jnrB+0];
 469             vdwjidx0C        = 2*vdwtype[jnrC+0];
 470             vdwjidx0D        = 2*vdwtype[jnrD+0];
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             r00              = _mm_mul_ps(rsq00,rinv00);
 477             r00              = _mm_andnot_ps(dummy_mask,r00);
 478
 479             /* Compute parameters for interactions between i and j atoms */
 480             qq00             = _mm_mul_ps(iq0,jq0);
 481             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 482                                          vdwparam+vdwioffset0+vdwjidx0B,
 483                                          vdwparam+vdwioffset0+vdwjidx0C,
 484                                          vdwparam+vdwioffset0+vdwjidx0D,
 485                                          &c6_00,&c12_00);
 486
 487             /* Calculate table index by multiplying r with table scale and truncate to integer */
 488             rt               = _mm_mul_ps(r00,vftabscale);
 489             vfitab           = _mm_cvttps_epi32(rt);
 490             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 491             vfitab           = _mm_slli_epi32(vfitab,3);
 492
 493             /* EWALD ELECTROSTATICS */
 494
 495             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 496             ewrt             = _mm_mul_ps(r00,ewtabscale);
 497             ewitab           = _mm_cvttps_epi32(ewrt);
 498             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 499             ewitab           = _mm_slli_epi32(ewitab,2);
 500             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 501             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 502             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 503             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 504             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 505             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 506             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 507             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 508             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 509
 510             /* CUBIC SPLINE TABLE DISPERSION */
 511             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 512             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 513             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 514             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 515             _MM_TRANSPOSE4_PS(Y,F,G,H);
 516             Heps             = _mm_mul_ps(vfeps,H);
 517             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 518             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 519             vvdw6            = _mm_mul_ps(c6_00,VV);
 520             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 521             fvdw6            = _mm_mul_ps(c6_00,FF);
 522
 523             /* CUBIC SPLINE TABLE REPULSION */
 524             vfitab           = _mm_add_epi32(vfitab,ifour);
 525             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 526             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 527             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 528             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 529             _MM_TRANSPOSE4_PS(Y,F,G,H);
 530             Heps             = _mm_mul_ps(vfeps,H);
 531             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 532             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 533             vvdw12           = _mm_mul_ps(c12_00,VV);
 534             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 535             fvdw12           = _mm_mul_ps(c12_00,FF);
 536             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 537             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 538
 539             /* Update potential sum for this i atom from the interaction with this j atom. */
 540             velec            = _mm_andnot_ps(dummy_mask,velec);
 541             velecsum         = _mm_add_ps(velecsum,velec);
 542             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 543             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 544
 545             fscal            = _mm_add_ps(felec,fvdw);
 546
 547             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 548
 549             /* Calculate temporary vectorial force */
 550             tx               = _mm_mul_ps(fscal,dx00);
 551             ty               = _mm_mul_ps(fscal,dy00);
 552             tz               = _mm_mul_ps(fscal,dz00);
 553
 554             /* Update vectorial force */
 555             fix0             = _mm_add_ps(fix0,tx);
 556             fiy0             = _mm_add_ps(fiy0,ty);
 557             fiz0             = _mm_add_ps(fiz0,tz);
 558
 559             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 560                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 561                                                    tx,ty,tz);
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             r10              = _mm_mul_ps(rsq10,rinv10);
 568             r10              = _mm_andnot_ps(dummy_mask,r10);
 569
 570             /* Compute parameters for interactions between i and j atoms */
 571             qq10             = _mm_mul_ps(iq1,jq0);
 572
 573             /* EWALD ELECTROSTATICS */
 574
 575             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 576             ewrt             = _mm_mul_ps(r10,ewtabscale);
 577             ewitab           = _mm_cvttps_epi32(ewrt);
 578             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 579             ewitab           = _mm_slli_epi32(ewitab,2);
 580             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 581             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 582             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 583             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 584             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 585             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 586             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 587             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 588             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 589
 590             /* Update potential sum for this i atom from the interaction with this j atom. */
 591             velec            = _mm_andnot_ps(dummy_mask,velec);
 592             velecsum         = _mm_add_ps(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm_mul_ps(fscal,dx10);
 600             ty               = _mm_mul_ps(fscal,dy10);
 601             tz               = _mm_mul_ps(fscal,dz10);
 602
 603             /* Update vectorial force */
 604             fix1             = _mm_add_ps(fix1,tx);
 605             fiy1             = _mm_add_ps(fiy1,ty);
 606             fiz1             = _mm_add_ps(fiz1,tz);
 607
 608             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 609                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 610                                                    tx,ty,tz);
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             r20              = _mm_mul_ps(rsq20,rinv20);
 617             r20              = _mm_andnot_ps(dummy_mask,r20);
 618
 619             /* Compute parameters for interactions between i and j atoms */
 620             qq20             = _mm_mul_ps(iq2,jq0);
 621
 622             /* EWALD ELECTROSTATICS */
 623
 624             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 625             ewrt             = _mm_mul_ps(r20,ewtabscale);
 626             ewitab           = _mm_cvttps_epi32(ewrt);
 627             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 628             ewitab           = _mm_slli_epi32(ewitab,2);
 629             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 630             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 631             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 632             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 633             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 634             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 635             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 636             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 637             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 638
 639             /* Update potential sum for this i atom from the interaction with this j atom. */
 640             velec            = _mm_andnot_ps(dummy_mask,velec);
 641             velecsum         = _mm_add_ps(velecsum,velec);
 642
 643             fscal            = felec;
 644
 645             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm_mul_ps(fscal,dx20);
 649             ty               = _mm_mul_ps(fscal,dy20);
 650             tz               = _mm_mul_ps(fscal,dz20);
 651
 652             /* Update vectorial force */
 653             fix2             = _mm_add_ps(fix2,tx);
 654             fiy2             = _mm_add_ps(fiy2,ty);
 655             fiz2             = _mm_add_ps(fiz2,tz);
 656
 657             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 658                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 659                                                    tx,ty,tz);
 660
 661             /* Inner loop uses 160 flops */
 662         }
 663
 664         /* End of innermost loop */
 665
 666         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 667                                               f+i_coord_offset,fshift+i_shift_offset);
 668
 669         ggid                        = gid[iidx];
 670         /* Update potential energies */
 671         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 672         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 673
 674         /* Increment number of inner iterations */
 675         inneriter                  += j_index_end - j_index_start;
 676
 677         /* Outer loop uses 29 flops */
 678     }
 679
 680     /* Increment number of outer iterations */
 681     outeriter        += nri;
 682
 683     /* Update outer/inner flops */
 684
 685     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*29 + inneriter*160);
 686 }
 687 /*
 688  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_single
 689  * Electrostatics interaction: Ewald
 690  * VdW interaction:            CubicSplineTable
 691  * Geometry:                   Water3-Particle
 692  * Calculate force/pot:        Force
 693  */
 694 void
 695 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_single
 696                     (t_nblist * gmx_restrict                nlist,
 697                      rvec * gmx_restrict                    xx,
 698                      rvec * gmx_restrict                    ff,
 699                      t_forcerec * gmx_restrict              fr,
 700                      t_mdatoms * gmx_restrict               mdatoms,
 701                      nb_kernel_data_t * gmx_restrict        kernel_data,
 702                      t_nrnb * gmx_restrict                  nrnb)
 703 {
 704     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 705      * just 0 for non-waters.
 706      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 707      * jnr indices corresponding to data put in the four positions in the SIMD register.
 708      */
 709     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 710     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 711     int              jnrA,jnrB,jnrC,jnrD;
 712     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 713     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 714     real             shX,shY,shZ,rcutoff_scalar;
 715     real             *shiftvec,*fshift,*x,*f;
 716     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 717     int              vdwioffset0;
 718     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 719     int              vdwioffset1;
 720     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 721     int              vdwioffset2;
 722     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 723     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 724     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 725     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 726     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 727     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 728     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 729     real             *charge;
 730     int              nvdwtype;
 731     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 732     int              *vdwtype;
 733     real             *vdwparam;
 734     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 735     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 736     __m128i          vfitab;
 737     __m128i          ifour       = _mm_set1_epi32(4);
 738     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 739     real             *vftab;
 740     __m128i          ewitab;
 741     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 742     real             *ewtab;
 743     __m128           dummy_mask,cutoff_mask;
 744     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 745     __m128           one     = _mm_set1_ps(1.0);
 746     __m128           two     = _mm_set1_ps(2.0);
 747     x                = xx[0];
 748     f                = ff[0];
 749
 750     nri              = nlist->nri;
 751     iinr             = nlist->iinr;
 752     jindex           = nlist->jindex;
 753     jjnr             = nlist->jjnr;
 754     shiftidx         = nlist->shift;
 755     gid              = nlist->gid;
 756     shiftvec         = fr->shift_vec[0];
 757     fshift           = fr->fshift[0];
 758     facel            = _mm_set1_ps(fr->epsfac);
 759     charge           = mdatoms->chargeA;
 760     nvdwtype         = fr->ntype;
 761     vdwparam         = fr->nbfp;
 762     vdwtype          = mdatoms->typeA;
 763
 764     vftab            = kernel_data->table_vdw->data;
 765     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 766
 767     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 768     ewtab            = fr->ic->tabq_coul_F;
 769     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 770     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 771
 772     /* Setup water-specific parameters */
 773     inr              = nlist->iinr[0];
 774     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 775     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 776     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 777     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 778
 779     /* Avoid stupid compiler warnings */
 780     jnrA = jnrB = jnrC = jnrD = 0;
 781     j_coord_offsetA = 0;
 782     j_coord_offsetB = 0;
 783     j_coord_offsetC = 0;
 784     j_coord_offsetD = 0;
 785
 786     outeriter        = 0;
 787     inneriter        = 0;
 788
 789     /* Start outer loop over neighborlists */
 790     for(iidx=0; iidx<nri; iidx++)
 791     {
 792         /* Load shift vector for this list */
 793         i_shift_offset   = DIM*shiftidx[iidx];
 794         shX              = shiftvec[i_shift_offset+XX];
 795         shY              = shiftvec[i_shift_offset+YY];
 796         shZ              = shiftvec[i_shift_offset+ZZ];
 797
 798         /* Load limits for loop over neighbors */
 799         j_index_start    = jindex[iidx];
 800         j_index_end      = jindex[iidx+1];
 801
 802         /* Get outer coordinate index */
 803         inr              = iinr[iidx];
 804         i_coord_offset   = DIM*inr;
 805
 806         /* Load i particle coords and add shift vector */
 807         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 808         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 809         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 810         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 811         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 812         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 813         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 814         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 815         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 816
 817         fix0             = _mm_setzero_ps();
 818         fiy0             = _mm_setzero_ps();
 819         fiz0             = _mm_setzero_ps();
 820         fix1             = _mm_setzero_ps();
 821         fiy1             = _mm_setzero_ps();
 822         fiz1             = _mm_setzero_ps();
 823         fix2             = _mm_setzero_ps();
 824         fiy2             = _mm_setzero_ps();
 825         fiz2             = _mm_setzero_ps();
 826
 827         /* Start inner kernel loop */
 828         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 829         {
 830
 831             /* Get j neighbor index, and coordinate index */
 832             jnrA             = jjnr[jidx];
 833             jnrB             = jjnr[jidx+1];
 834             jnrC             = jjnr[jidx+2];
 835             jnrD             = jjnr[jidx+3];
 836
 837             j_coord_offsetA  = DIM*jnrA;
 838             j_coord_offsetB  = DIM*jnrB;
 839             j_coord_offsetC  = DIM*jnrC;
 840             j_coord_offsetD  = DIM*jnrD;
 841
 842             /* load j atom coordinates */
 843             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 844                                               x+j_coord_offsetC,x+j_coord_offsetD,
 845                                               &jx0,&jy0,&jz0);
 846
 847             /* Calculate displacement vector */
 848             dx00             = _mm_sub_ps(ix0,jx0);
 849             dy00             = _mm_sub_ps(iy0,jy0);
 850             dz00             = _mm_sub_ps(iz0,jz0);
 851             dx10             = _mm_sub_ps(ix1,jx0);
 852             dy10             = _mm_sub_ps(iy1,jy0);
 853             dz10             = _mm_sub_ps(iz1,jz0);
 854             dx20             = _mm_sub_ps(ix2,jx0);
 855             dy20             = _mm_sub_ps(iy2,jy0);
 856             dz20             = _mm_sub_ps(iz2,jz0);
 857
 858             /* Calculate squared distance and things based on it */
 859             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 860             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 861             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 862
 863             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 864             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 865             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 866
 867             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 868             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 869             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 870
 871             /* Load parameters for j particles */
 872             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 873                                                               charge+jnrC+0,charge+jnrD+0);
 874             vdwjidx0A        = 2*vdwtype[jnrA+0];
 875             vdwjidx0B        = 2*vdwtype[jnrB+0];
 876             vdwjidx0C        = 2*vdwtype[jnrC+0];
 877             vdwjidx0D        = 2*vdwtype[jnrD+0];
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             r00              = _mm_mul_ps(rsq00,rinv00);
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             qq00             = _mm_mul_ps(iq0,jq0);
 887             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 888                                          vdwparam+vdwioffset0+vdwjidx0B,
 889                                          vdwparam+vdwioffset0+vdwjidx0C,
 890                                          vdwparam+vdwioffset0+vdwjidx0D,
 891                                          &c6_00,&c12_00);
 892
 893             /* Calculate table index by multiplying r with table scale and truncate to integer */
 894             rt               = _mm_mul_ps(r00,vftabscale);
 895             vfitab           = _mm_cvttps_epi32(rt);
 896             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 897             vfitab           = _mm_slli_epi32(vfitab,3);
 898
 899             /* EWALD ELECTROSTATICS */
 900
 901             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 902             ewrt             = _mm_mul_ps(r00,ewtabscale);
 903             ewitab           = _mm_cvttps_epi32(ewrt);
 904             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 905             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 906                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 907                                          &ewtabF,&ewtabFn);
 908             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 909             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 910
 911             /* CUBIC SPLINE TABLE DISPERSION */
 912             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 913             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 914             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 915             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 916             _MM_TRANSPOSE4_PS(Y,F,G,H);
 917             Heps             = _mm_mul_ps(vfeps,H);
 918             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 919             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 920             fvdw6            = _mm_mul_ps(c6_00,FF);
 921
 922             /* CUBIC SPLINE TABLE REPULSION */
 923             vfitab           = _mm_add_epi32(vfitab,ifour);
 924             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 925             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 926             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 927             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 928             _MM_TRANSPOSE4_PS(Y,F,G,H);
 929             Heps             = _mm_mul_ps(vfeps,H);
 930             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 931             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 932             fvdw12           = _mm_mul_ps(c12_00,FF);
 933             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 934
 935             fscal            = _mm_add_ps(felec,fvdw);
 936
 937             /* Calculate temporary vectorial force */
 938             tx               = _mm_mul_ps(fscal,dx00);
 939             ty               = _mm_mul_ps(fscal,dy00);
 940             tz               = _mm_mul_ps(fscal,dz00);
 941
 942             /* Update vectorial force */
 943             fix0             = _mm_add_ps(fix0,tx);
 944             fiy0             = _mm_add_ps(fiy0,ty);
 945             fiz0             = _mm_add_ps(fiz0,tz);
 946
 947             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 948                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 949                                                    tx,ty,tz);
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             r10              = _mm_mul_ps(rsq10,rinv10);
 956
 957             /* Compute parameters for interactions between i and j atoms */
 958             qq10             = _mm_mul_ps(iq1,jq0);
 959
 960             /* EWALD ELECTROSTATICS */
 961
 962             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 963             ewrt             = _mm_mul_ps(r10,ewtabscale);
 964             ewitab           = _mm_cvttps_epi32(ewrt);
 965             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 966             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 967                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 968                                          &ewtabF,&ewtabFn);
 969             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 970             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 971
 972             fscal            = felec;
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm_mul_ps(fscal,dx10);
 976             ty               = _mm_mul_ps(fscal,dy10);
 977             tz               = _mm_mul_ps(fscal,dz10);
 978
 979             /* Update vectorial force */
 980             fix1             = _mm_add_ps(fix1,tx);
 981             fiy1             = _mm_add_ps(fiy1,ty);
 982             fiz1             = _mm_add_ps(fiz1,tz);
 983
 984             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 985                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 986                                                    tx,ty,tz);
 987
 988             /**************************
 989              * CALCULATE INTERACTIONS *
 990              **************************/
 991
 992             r20              = _mm_mul_ps(rsq20,rinv20);
 993
 994             /* Compute parameters for interactions between i and j atoms */
 995             qq20             = _mm_mul_ps(iq2,jq0);
 996
 997             /* EWALD ELECTROSTATICS */
 998
 999             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1000             ewrt             = _mm_mul_ps(r20,ewtabscale);
1001             ewitab           = _mm_cvttps_epi32(ewrt);
1002             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1003             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1004                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1005                                          &ewtabF,&ewtabFn);
1006             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1007             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1008
1009             fscal            = felec;
1010
1011             /* Calculate temporary vectorial force */
1012             tx               = _mm_mul_ps(fscal,dx20);
1013             ty               = _mm_mul_ps(fscal,dy20);
1014             tz               = _mm_mul_ps(fscal,dz20);
1015
1016             /* Update vectorial force */
1017             fix2             = _mm_add_ps(fix2,tx);
1018             fiy2             = _mm_add_ps(fiy2,ty);
1019             fiz2             = _mm_add_ps(fiz2,tz);
1020
1021             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1022                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1023                                                    tx,ty,tz);
1024
1025             /* Inner loop uses 134 flops */
1026         }
1027
1028         if(jidx<j_index_end)
1029         {
1030
1031             /* Get j neighbor index, and coordinate index */
1032             jnrA             = jjnr[jidx];
1033             jnrB             = jjnr[jidx+1];
1034             jnrC             = jjnr[jidx+2];
1035             jnrD             = jjnr[jidx+3];
1036
1037             /* Sign of each element will be negative for non-real atoms.
1038              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1039              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1040              */
1041             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1042             jnrA       = (jnrA>=0) ? jnrA : 0;
1043             jnrB       = (jnrB>=0) ? jnrB : 0;
1044             jnrC       = (jnrC>=0) ? jnrC : 0;
1045             jnrD       = (jnrD>=0) ? jnrD : 0;
1046
1047             j_coord_offsetA  = DIM*jnrA;
1048             j_coord_offsetB  = DIM*jnrB;
1049             j_coord_offsetC  = DIM*jnrC;
1050             j_coord_offsetD  = DIM*jnrD;
1051
1052             /* load j atom coordinates */
1053             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1054                                               x+j_coord_offsetC,x+j_coord_offsetD,
1055                                               &jx0,&jy0,&jz0);
1056
1057             /* Calculate displacement vector */
1058             dx00             = _mm_sub_ps(ix0,jx0);
1059             dy00             = _mm_sub_ps(iy0,jy0);
1060             dz00             = _mm_sub_ps(iz0,jz0);
1061             dx10             = _mm_sub_ps(ix1,jx0);
1062             dy10             = _mm_sub_ps(iy1,jy0);
1063             dz10             = _mm_sub_ps(iz1,jz0);
1064             dx20             = _mm_sub_ps(ix2,jx0);
1065             dy20             = _mm_sub_ps(iy2,jy0);
1066             dz20             = _mm_sub_ps(iz2,jz0);
1067
1068             /* Calculate squared distance and things based on it */
1069             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1070             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1071             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1072
1073             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1074             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1075             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1076
1077             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1078             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1079             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1080
1081             /* Load parameters for j particles */
1082             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1083                                                               charge+jnrC+0,charge+jnrD+0);
1084             vdwjidx0A        = 2*vdwtype[jnrA+0];
1085             vdwjidx0B        = 2*vdwtype[jnrB+0];
1086             vdwjidx0C        = 2*vdwtype[jnrC+0];
1087             vdwjidx0D        = 2*vdwtype[jnrD+0];
1088
1089             /**************************
1090              * CALCULATE INTERACTIONS *
1091              **************************/
1092
1093             r00              = _mm_mul_ps(rsq00,rinv00);
1094             r00              = _mm_andnot_ps(dummy_mask,r00);
1095
1096             /* Compute parameters for interactions between i and j atoms */
1097             qq00             = _mm_mul_ps(iq0,jq0);
1098             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1099                                          vdwparam+vdwioffset0+vdwjidx0B,
1100                                          vdwparam+vdwioffset0+vdwjidx0C,
1101                                          vdwparam+vdwioffset0+vdwjidx0D,
1102                                          &c6_00,&c12_00);
1103
1104             /* Calculate table index by multiplying r with table scale and truncate to integer */
1105             rt               = _mm_mul_ps(r00,vftabscale);
1106             vfitab           = _mm_cvttps_epi32(rt);
1107             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1108             vfitab           = _mm_slli_epi32(vfitab,3);
1109
1110             /* EWALD ELECTROSTATICS */
1111
1112             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1113             ewrt             = _mm_mul_ps(r00,ewtabscale);
1114             ewitab           = _mm_cvttps_epi32(ewrt);
1115             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1116             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1117                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1118                                          &ewtabF,&ewtabFn);
1119             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1120             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1121
1122             /* CUBIC SPLINE TABLE DISPERSION */
1123             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1124             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1125             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1126             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1127             _MM_TRANSPOSE4_PS(Y,F,G,H);
1128             Heps             = _mm_mul_ps(vfeps,H);
1129             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1130             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1131             fvdw6            = _mm_mul_ps(c6_00,FF);
1132
1133             /* CUBIC SPLINE TABLE REPULSION */
1134             vfitab           = _mm_add_epi32(vfitab,ifour);
1135             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1136             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1137             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1138             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1139             _MM_TRANSPOSE4_PS(Y,F,G,H);
1140             Heps             = _mm_mul_ps(vfeps,H);
1141             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1142             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1143             fvdw12           = _mm_mul_ps(c12_00,FF);
1144             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1145
1146             fscal            = _mm_add_ps(felec,fvdw);
1147
1148             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1149
1150             /* Calculate temporary vectorial force */
1151             tx               = _mm_mul_ps(fscal,dx00);
1152             ty               = _mm_mul_ps(fscal,dy00);
1153             tz               = _mm_mul_ps(fscal,dz00);
1154
1155             /* Update vectorial force */
1156             fix0             = _mm_add_ps(fix0,tx);
1157             fiy0             = _mm_add_ps(fiy0,ty);
1158             fiz0             = _mm_add_ps(fiz0,tz);
1159
1160             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1161                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1162                                                    tx,ty,tz);
1163
1164             /**************************
1165              * CALCULATE INTERACTIONS *
1166              **************************/
1167
1168             r10              = _mm_mul_ps(rsq10,rinv10);
1169             r10              = _mm_andnot_ps(dummy_mask,r10);
1170
1171             /* Compute parameters for interactions between i and j atoms */
1172             qq10             = _mm_mul_ps(iq1,jq0);
1173
1174             /* EWALD ELECTROSTATICS */
1175
1176             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1177             ewrt             = _mm_mul_ps(r10,ewtabscale);
1178             ewitab           = _mm_cvttps_epi32(ewrt);
1179             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1180             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1181                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1182                                          &ewtabF,&ewtabFn);
1183             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1184             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1185
1186             fscal            = felec;
1187
1188             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1189
1190             /* Calculate temporary vectorial force */
1191             tx               = _mm_mul_ps(fscal,dx10);
1192             ty               = _mm_mul_ps(fscal,dy10);
1193             tz               = _mm_mul_ps(fscal,dz10);
1194
1195             /* Update vectorial force */
1196             fix1             = _mm_add_ps(fix1,tx);
1197             fiy1             = _mm_add_ps(fiy1,ty);
1198             fiz1             = _mm_add_ps(fiz1,tz);
1199
1200             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1201                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1202                                                    tx,ty,tz);
1203
1204             /**************************
1205              * CALCULATE INTERACTIONS *
1206              **************************/
1207
1208             r20              = _mm_mul_ps(rsq20,rinv20);
1209             r20              = _mm_andnot_ps(dummy_mask,r20);
1210
1211             /* Compute parameters for interactions between i and j atoms */
1212             qq20             = _mm_mul_ps(iq2,jq0);
1213
1214             /* EWALD ELECTROSTATICS */
1215
1216             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1217             ewrt             = _mm_mul_ps(r20,ewtabscale);
1218             ewitab           = _mm_cvttps_epi32(ewrt);
1219             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1220             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1221                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1222                                          &ewtabF,&ewtabFn);
1223             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1224             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1225
1226             fscal            = felec;
1227
1228             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1229
1230             /* Calculate temporary vectorial force */
1231             tx               = _mm_mul_ps(fscal,dx20);
1232             ty               = _mm_mul_ps(fscal,dy20);
1233             tz               = _mm_mul_ps(fscal,dz20);
1234
1235             /* Update vectorial force */
1236             fix2             = _mm_add_ps(fix2,tx);
1237             fiy2             = _mm_add_ps(fiy2,ty);
1238             fiz2             = _mm_add_ps(fiz2,tz);
1239
1240             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1241                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1242                                                    tx,ty,tz);
1243
1244             /* Inner loop uses 137 flops */
1245         }
1246
1247         /* End of innermost loop */
1248
1249         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1250                                               f+i_coord_offset,fshift+i_shift_offset);
1251
1252         /* Increment number of inner iterations */
1253         inneriter                  += j_index_end - j_index_start;
1254
1255         /* Outer loop uses 27 flops */
1256     }
1257
1258     /* Increment number of outer iterations */
1259     outeriter        += nri;
1260
1261     /* Update outer/inner flops */
1262
1263     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*27 + inneriter*137);
1264 }