src/gromacs/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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  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     __m128i          ewitab;
  93     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     real             *ewtab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 118
 119     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 127     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 128     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 129     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 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
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_ps();
 176         vvdwsum          = _mm_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               x+j_coord_offsetC,x+j_coord_offsetD,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_ps(ix0,jx0);
 199             dy00             = _mm_sub_ps(iy0,jy0);
 200             dz00             = _mm_sub_ps(iz0,jz0);
 201             dx10             = _mm_sub_ps(ix1,jx0);
 202             dy10             = _mm_sub_ps(iy1,jy0);
 203             dz10             = _mm_sub_ps(iz1,jz0);
 204             dx20             = _mm_sub_ps(ix2,jx0);
 205             dy20             = _mm_sub_ps(iy2,jy0);
 206             dz20             = _mm_sub_ps(iz2,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 210             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 211             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 212
 213             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 214             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 215             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 216
 217             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 218             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 219             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 223                                                               charge+jnrC+0,charge+jnrD+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             r00              = _mm_mul_ps(rsq00,rinv00);
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             qq00             = _mm_mul_ps(iq0,jq0);
 237             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 238                                          vdwparam+vdwioffset0+vdwjidx0B,
 239                                          vdwparam+vdwioffset0+vdwjidx0C,
 240                                          vdwparam+vdwioffset0+vdwjidx0D,
 241                                          &c6_00,&c12_00);
 242
 243             /* Calculate table index by multiplying r with table scale and truncate to integer */
 244             rt               = _mm_mul_ps(r00,vftabscale);
 245             vfitab           = _mm_cvttps_epi32(rt);
 246             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 247             vfitab           = _mm_slli_epi32(vfitab,3);
 248
 249             /* EWALD ELECTROSTATICS */
 250
 251             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 252             ewrt             = _mm_mul_ps(r00,ewtabscale);
 253             ewitab           = _mm_cvttps_epi32(ewrt);
 254             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 255             ewitab           = _mm_slli_epi32(ewitab,2);
 256             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 257             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 258             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 259             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 260             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 261             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 262             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 263             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 264             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 265
 266             /* CUBIC SPLINE TABLE DISPERSION */
 267             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 268             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 269             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 270             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 271             _MM_TRANSPOSE4_PS(Y,F,G,H);
 272             Heps             = _mm_mul_ps(vfeps,H);
 273             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 274             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 275             vvdw6            = _mm_mul_ps(c6_00,VV);
 276             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 277             fvdw6            = _mm_mul_ps(c6_00,FF);
 278
 279             /* CUBIC SPLINE TABLE REPULSION */
 280             vfitab           = _mm_add_epi32(vfitab,ifour);
 281             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 282             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 283             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 284             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 285             _MM_TRANSPOSE4_PS(Y,F,G,H);
 286             Heps             = _mm_mul_ps(vfeps,H);
 287             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 288             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 289             vvdw12           = _mm_mul_ps(c12_00,VV);
 290             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 291             fvdw12           = _mm_mul_ps(c12_00,FF);
 292             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 293             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velecsum         = _mm_add_ps(velecsum,velec);
 297             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 298
 299             fscal            = _mm_add_ps(felec,fvdw);
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm_mul_ps(fscal,dx00);
 303             ty               = _mm_mul_ps(fscal,dy00);
 304             tz               = _mm_mul_ps(fscal,dz00);
 305
 306             /* Update vectorial force */
 307             fix0             = _mm_add_ps(fix0,tx);
 308             fiy0             = _mm_add_ps(fiy0,ty);
 309             fiz0             = _mm_add_ps(fiz0,tz);
 310
 311             fjptrA             = f+j_coord_offsetA;
 312             fjptrB             = f+j_coord_offsetB;
 313             fjptrC             = f+j_coord_offsetC;
 314             fjptrD             = f+j_coord_offsetD;
 315             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             r10              = _mm_mul_ps(rsq10,rinv10);
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm_mul_ps(iq1,jq0);
 325
 326             /* EWALD ELECTROSTATICS */
 327
 328             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 329             ewrt             = _mm_mul_ps(r10,ewtabscale);
 330             ewitab           = _mm_cvttps_epi32(ewrt);
 331             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 332             ewitab           = _mm_slli_epi32(ewitab,2);
 333             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 334             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 335             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 336             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 337             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 338             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 339             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 340             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 341             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm_add_ps(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_ps(fscal,dx10);
 350             ty               = _mm_mul_ps(fscal,dy10);
 351             tz               = _mm_mul_ps(fscal,dz10);
 352
 353             /* Update vectorial force */
 354             fix1             = _mm_add_ps(fix1,tx);
 355             fiy1             = _mm_add_ps(fiy1,ty);
 356             fiz1             = _mm_add_ps(fiz1,tz);
 357
 358             fjptrA             = f+j_coord_offsetA;
 359             fjptrB             = f+j_coord_offsetB;
 360             fjptrC             = f+j_coord_offsetC;
 361             fjptrD             = f+j_coord_offsetD;
 362             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             r20              = _mm_mul_ps(rsq20,rinv20);
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq20             = _mm_mul_ps(iq2,jq0);
 372
 373             /* EWALD ELECTROSTATICS */
 374
 375             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 376             ewrt             = _mm_mul_ps(r20,ewtabscale);
 377             ewitab           = _mm_cvttps_epi32(ewrt);
 378             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 379             ewitab           = _mm_slli_epi32(ewitab,2);
 380             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 381             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 382             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 383             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 384             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 385             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 386             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 387             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 388             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velecsum         = _mm_add_ps(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm_mul_ps(fscal,dx20);
 397             ty               = _mm_mul_ps(fscal,dy20);
 398             tz               = _mm_mul_ps(fscal,dz20);
 399
 400             /* Update vectorial force */
 401             fix2             = _mm_add_ps(fix2,tx);
 402             fiy2             = _mm_add_ps(fiy2,ty);
 403             fiz2             = _mm_add_ps(fiz2,tz);
 404
 405             fjptrA             = f+j_coord_offsetA;
 406             fjptrB             = f+j_coord_offsetB;
 407             fjptrC             = f+j_coord_offsetC;
 408             fjptrD             = f+j_coord_offsetD;
 409             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 410
 411             /* Inner loop uses 157 flops */
 412         }
 413
 414         if(jidx<j_index_end)
 415         {
 416
 417             /* Get j neighbor index, and coordinate index */
 418             jnrlistA         = jjnr[jidx];
 419             jnrlistB         = jjnr[jidx+1];
 420             jnrlistC         = jjnr[jidx+2];
 421             jnrlistD         = jjnr[jidx+3];
 422             /* Sign of each element will be negative for non-real atoms.
 423              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 424              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 425              */
 426             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 427             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 428             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 429             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 430             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 431             j_coord_offsetA  = DIM*jnrA;
 432             j_coord_offsetB  = DIM*jnrB;
 433             j_coord_offsetC  = DIM*jnrC;
 434             j_coord_offsetD  = DIM*jnrD;
 435
 436             /* load j atom coordinates */
 437             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 438                                               x+j_coord_offsetC,x+j_coord_offsetD,
 439                                               &jx0,&jy0,&jz0);
 440
 441             /* Calculate displacement vector */
 442             dx00             = _mm_sub_ps(ix0,jx0);
 443             dy00             = _mm_sub_ps(iy0,jy0);
 444             dz00             = _mm_sub_ps(iz0,jz0);
 445             dx10             = _mm_sub_ps(ix1,jx0);
 446             dy10             = _mm_sub_ps(iy1,jy0);
 447             dz10             = _mm_sub_ps(iz1,jz0);
 448             dx20             = _mm_sub_ps(ix2,jx0);
 449             dy20             = _mm_sub_ps(iy2,jy0);
 450             dz20             = _mm_sub_ps(iz2,jz0);
 451
 452             /* Calculate squared distance and things based on it */
 453             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 454             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 455             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 456
 457             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 458             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 459             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 460
 461             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 462             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 463             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 464
 465             /* Load parameters for j particles */
 466             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 467                                                               charge+jnrC+0,charge+jnrD+0);
 468             vdwjidx0A        = 2*vdwtype[jnrA+0];
 469             vdwjidx0B        = 2*vdwtype[jnrB+0];
 470             vdwjidx0C        = 2*vdwtype[jnrC+0];
 471             vdwjidx0D        = 2*vdwtype[jnrD+0];
 472
 473             /**************************
 474              * CALCULATE INTERACTIONS *
 475              **************************/
 476
 477             r00              = _mm_mul_ps(rsq00,rinv00);
 478             r00              = _mm_andnot_ps(dummy_mask,r00);
 479
 480             /* Compute parameters for interactions between i and j atoms */
 481             qq00             = _mm_mul_ps(iq0,jq0);
 482             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 483                                          vdwparam+vdwioffset0+vdwjidx0B,
 484                                          vdwparam+vdwioffset0+vdwjidx0C,
 485                                          vdwparam+vdwioffset0+vdwjidx0D,
 486                                          &c6_00,&c12_00);
 487
 488             /* Calculate table index by multiplying r with table scale and truncate to integer */
 489             rt               = _mm_mul_ps(r00,vftabscale);
 490             vfitab           = _mm_cvttps_epi32(rt);
 491             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 492             vfitab           = _mm_slli_epi32(vfitab,3);
 493
 494             /* EWALD ELECTROSTATICS */
 495
 496             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 497             ewrt             = _mm_mul_ps(r00,ewtabscale);
 498             ewitab           = _mm_cvttps_epi32(ewrt);
 499             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 500             ewitab           = _mm_slli_epi32(ewitab,2);
 501             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 502             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 503             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 504             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 505             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 506             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 507             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 508             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 509             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 510
 511             /* CUBIC SPLINE TABLE DISPERSION */
 512             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 513             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 514             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 515             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 516             _MM_TRANSPOSE4_PS(Y,F,G,H);
 517             Heps             = _mm_mul_ps(vfeps,H);
 518             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 519             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 520             vvdw6            = _mm_mul_ps(c6_00,VV);
 521             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 522             fvdw6            = _mm_mul_ps(c6_00,FF);
 523
 524             /* CUBIC SPLINE TABLE REPULSION */
 525             vfitab           = _mm_add_epi32(vfitab,ifour);
 526             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 527             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 528             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 529             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 530             _MM_TRANSPOSE4_PS(Y,F,G,H);
 531             Heps             = _mm_mul_ps(vfeps,H);
 532             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 533             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 534             vvdw12           = _mm_mul_ps(c12_00,VV);
 535             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 536             fvdw12           = _mm_mul_ps(c12_00,FF);
 537             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 538             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             velec            = _mm_andnot_ps(dummy_mask,velec);
 542             velecsum         = _mm_add_ps(velecsum,velec);
 543             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 544             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 545
 546             fscal            = _mm_add_ps(felec,fvdw);
 547
 548             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 549
 550             /* Calculate temporary vectorial force */
 551             tx               = _mm_mul_ps(fscal,dx00);
 552             ty               = _mm_mul_ps(fscal,dy00);
 553             tz               = _mm_mul_ps(fscal,dz00);
 554
 555             /* Update vectorial force */
 556             fix0             = _mm_add_ps(fix0,tx);
 557             fiy0             = _mm_add_ps(fiy0,ty);
 558             fiz0             = _mm_add_ps(fiz0,tz);
 559
 560             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 561             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 562             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 563             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 564             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             r10              = _mm_mul_ps(rsq10,rinv10);
 571             r10              = _mm_andnot_ps(dummy_mask,r10);
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             qq10             = _mm_mul_ps(iq1,jq0);
 575
 576             /* EWALD ELECTROSTATICS */
 577
 578             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 579             ewrt             = _mm_mul_ps(r10,ewtabscale);
 580             ewitab           = _mm_cvttps_epi32(ewrt);
 581             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 582             ewitab           = _mm_slli_epi32(ewitab,2);
 583             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 584             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 585             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 586             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 587             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 588             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 589             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 590             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 591             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm_andnot_ps(dummy_mask,velec);
 595             velecsum         = _mm_add_ps(velecsum,velec);
 596
 597             fscal            = felec;
 598
 599             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm_mul_ps(fscal,dx10);
 603             ty               = _mm_mul_ps(fscal,dy10);
 604             tz               = _mm_mul_ps(fscal,dz10);
 605
 606             /* Update vectorial force */
 607             fix1             = _mm_add_ps(fix1,tx);
 608             fiy1             = _mm_add_ps(fiy1,ty);
 609             fiz1             = _mm_add_ps(fiz1,tz);
 610
 611             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 612             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 613             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 614             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 615             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             r20              = _mm_mul_ps(rsq20,rinv20);
 622             r20              = _mm_andnot_ps(dummy_mask,r20);
 623
 624             /* Compute parameters for interactions between i and j atoms */
 625             qq20             = _mm_mul_ps(iq2,jq0);
 626
 627             /* EWALD ELECTROSTATICS */
 628
 629             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 630             ewrt             = _mm_mul_ps(r20,ewtabscale);
 631             ewitab           = _mm_cvttps_epi32(ewrt);
 632             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 633             ewitab           = _mm_slli_epi32(ewitab,2);
 634             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 635             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 636             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 637             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 638             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 639             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 640             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 641             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 642             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 643
 644             /* Update potential sum for this i atom from the interaction with this j atom. */
 645             velec            = _mm_andnot_ps(dummy_mask,velec);
 646             velecsum         = _mm_add_ps(velecsum,velec);
 647
 648             fscal            = felec;
 649
 650             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 651
 652             /* Calculate temporary vectorial force */
 653             tx               = _mm_mul_ps(fscal,dx20);
 654             ty               = _mm_mul_ps(fscal,dy20);
 655             tz               = _mm_mul_ps(fscal,dz20);
 656
 657             /* Update vectorial force */
 658             fix2             = _mm_add_ps(fix2,tx);
 659             fiy2             = _mm_add_ps(fiy2,ty);
 660             fiz2             = _mm_add_ps(fiz2,tz);
 661
 662             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 663             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 664             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 665             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 666             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 667
 668             /* Inner loop uses 160 flops */
 669         }
 670
 671         /* End of innermost loop */
 672
 673         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 674                                               f+i_coord_offset,fshift+i_shift_offset);
 675
 676         ggid                        = gid[iidx];
 677         /* Update potential energies */
 678         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 679         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 680
 681         /* Increment number of inner iterations */
 682         inneriter                  += j_index_end - j_index_start;
 683
 684         /* Outer loop uses 20 flops */
 685     }
 686
 687     /* Increment number of outer iterations */
 688     outeriter        += nri;
 689
 690     /* Update outer/inner flops */
 691
 692     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*160);
 693 }
 694 /*
 695  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_single
 696  * Electrostatics interaction: Ewald
 697  * VdW interaction:            CubicSplineTable
 698  * Geometry:                   Water3-Particle
 699  * Calculate force/pot:        Force
 700  */
 701 void
 702 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_single
 703                     (t_nblist * gmx_restrict                nlist,
 704                      rvec * gmx_restrict                    xx,
 705                      rvec * gmx_restrict                    ff,
 706                      t_forcerec * gmx_restrict              fr,
 707                      t_mdatoms * gmx_restrict               mdatoms,
 708                      nb_kernel_data_t * gmx_restrict        kernel_data,
 709                      t_nrnb * gmx_restrict                  nrnb)
 710 {
 711     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 712      * just 0 for non-waters.
 713      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 714      * jnr indices corresponding to data put in the four positions in the SIMD register.
 715      */
 716     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 717     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 718     int              jnrA,jnrB,jnrC,jnrD;
 719     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 720     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 721     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 722     real             rcutoff_scalar;
 723     real             *shiftvec,*fshift,*x,*f;
 724     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 725     real             scratch[4*DIM];
 726     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 727     int              vdwioffset0;
 728     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 729     int              vdwioffset1;
 730     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 731     int              vdwioffset2;
 732     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 733     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 734     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 735     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 736     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 737     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 738     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 739     real             *charge;
 740     int              nvdwtype;
 741     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 742     int              *vdwtype;
 743     real             *vdwparam;
 744     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 745     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 746     __m128i          vfitab;
 747     __m128i          ifour       = _mm_set1_epi32(4);
 748     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 749     real             *vftab;
 750     __m128i          ewitab;
 751     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 752     real             *ewtab;
 753     __m128           dummy_mask,cutoff_mask;
 754     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 755     __m128           one     = _mm_set1_ps(1.0);
 756     __m128           two     = _mm_set1_ps(2.0);
 757     x                = xx[0];
 758     f                = ff[0];
 759
 760     nri              = nlist->nri;
 761     iinr             = nlist->iinr;
 762     jindex           = nlist->jindex;
 763     jjnr             = nlist->jjnr;
 764     shiftidx         = nlist->shift;
 765     gid              = nlist->gid;
 766     shiftvec         = fr->shift_vec[0];
 767     fshift           = fr->fshift[0];
 768     facel            = _mm_set1_ps(fr->epsfac);
 769     charge           = mdatoms->chargeA;
 770     nvdwtype         = fr->ntype;
 771     vdwparam         = fr->nbfp;
 772     vdwtype          = mdatoms->typeA;
 773
 774     vftab            = kernel_data->table_vdw->data;
 775     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 776
 777     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 778     ewtab            = fr->ic->tabq_coul_F;
 779     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 780     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 781
 782     /* Setup water-specific parameters */
 783     inr              = nlist->iinr[0];
 784     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 785     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 786     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 787     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 788
 789     /* Avoid stupid compiler warnings */
 790     jnrA = jnrB = jnrC = jnrD = 0;
 791     j_coord_offsetA = 0;
 792     j_coord_offsetB = 0;
 793     j_coord_offsetC = 0;
 794     j_coord_offsetD = 0;
 795
 796     outeriter        = 0;
 797     inneriter        = 0;
 798
 799     for(iidx=0;iidx<4*DIM;iidx++)
 800     {
 801         scratch[iidx] = 0.0;
 802     }
 803
 804     /* Start outer loop over neighborlists */
 805     for(iidx=0; iidx<nri; iidx++)
 806     {
 807         /* Load shift vector for this list */
 808         i_shift_offset   = DIM*shiftidx[iidx];
 809
 810         /* Load limits for loop over neighbors */
 811         j_index_start    = jindex[iidx];
 812         j_index_end      = jindex[iidx+1];
 813
 814         /* Get outer coordinate index */
 815         inr              = iinr[iidx];
 816         i_coord_offset   = DIM*inr;
 817
 818         /* Load i particle coords and add shift vector */
 819         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 820                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 821
 822         fix0             = _mm_setzero_ps();
 823         fiy0             = _mm_setzero_ps();
 824         fiz0             = _mm_setzero_ps();
 825         fix1             = _mm_setzero_ps();
 826         fiy1             = _mm_setzero_ps();
 827         fiz1             = _mm_setzero_ps();
 828         fix2             = _mm_setzero_ps();
 829         fiy2             = _mm_setzero_ps();
 830         fiz2             = _mm_setzero_ps();
 831
 832         /* Start inner kernel loop */
 833         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 834         {
 835
 836             /* Get j neighbor index, and coordinate index */
 837             jnrA             = jjnr[jidx];
 838             jnrB             = jjnr[jidx+1];
 839             jnrC             = jjnr[jidx+2];
 840             jnrD             = jjnr[jidx+3];
 841             j_coord_offsetA  = DIM*jnrA;
 842             j_coord_offsetB  = DIM*jnrB;
 843             j_coord_offsetC  = DIM*jnrC;
 844             j_coord_offsetD  = DIM*jnrD;
 845
 846             /* load j atom coordinates */
 847             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 848                                               x+j_coord_offsetC,x+j_coord_offsetD,
 849                                               &jx0,&jy0,&jz0);
 850
 851             /* Calculate displacement vector */
 852             dx00             = _mm_sub_ps(ix0,jx0);
 853             dy00             = _mm_sub_ps(iy0,jy0);
 854             dz00             = _mm_sub_ps(iz0,jz0);
 855             dx10             = _mm_sub_ps(ix1,jx0);
 856             dy10             = _mm_sub_ps(iy1,jy0);
 857             dz10             = _mm_sub_ps(iz1,jz0);
 858             dx20             = _mm_sub_ps(ix2,jx0);
 859             dy20             = _mm_sub_ps(iy2,jy0);
 860             dz20             = _mm_sub_ps(iz2,jz0);
 861
 862             /* Calculate squared distance and things based on it */
 863             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 864             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 865             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 866
 867             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 868             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 869             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 870
 871             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 872             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 873             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 874
 875             /* Load parameters for j particles */
 876             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 877                                                               charge+jnrC+0,charge+jnrD+0);
 878             vdwjidx0A        = 2*vdwtype[jnrA+0];
 879             vdwjidx0B        = 2*vdwtype[jnrB+0];
 880             vdwjidx0C        = 2*vdwtype[jnrC+0];
 881             vdwjidx0D        = 2*vdwtype[jnrD+0];
 882
 883             /**************************
 884              * CALCULATE INTERACTIONS *
 885              **************************/
 886
 887             r00              = _mm_mul_ps(rsq00,rinv00);
 888
 889             /* Compute parameters for interactions between i and j atoms */
 890             qq00             = _mm_mul_ps(iq0,jq0);
 891             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 892                                          vdwparam+vdwioffset0+vdwjidx0B,
 893                                          vdwparam+vdwioffset0+vdwjidx0C,
 894                                          vdwparam+vdwioffset0+vdwjidx0D,
 895                                          &c6_00,&c12_00);
 896
 897             /* Calculate table index by multiplying r with table scale and truncate to integer */
 898             rt               = _mm_mul_ps(r00,vftabscale);
 899             vfitab           = _mm_cvttps_epi32(rt);
 900             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 901             vfitab           = _mm_slli_epi32(vfitab,3);
 902
 903             /* EWALD ELECTROSTATICS */
 904
 905             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 906             ewrt             = _mm_mul_ps(r00,ewtabscale);
 907             ewitab           = _mm_cvttps_epi32(ewrt);
 908             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 909             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 910                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 911                                          &ewtabF,&ewtabFn);
 912             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 913             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 914
 915             /* CUBIC SPLINE TABLE DISPERSION */
 916             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 917             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 918             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 919             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 920             _MM_TRANSPOSE4_PS(Y,F,G,H);
 921             Heps             = _mm_mul_ps(vfeps,H);
 922             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 923             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 924             fvdw6            = _mm_mul_ps(c6_00,FF);
 925
 926             /* CUBIC SPLINE TABLE REPULSION */
 927             vfitab           = _mm_add_epi32(vfitab,ifour);
 928             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 929             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 930             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 931             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 932             _MM_TRANSPOSE4_PS(Y,F,G,H);
 933             Heps             = _mm_mul_ps(vfeps,H);
 934             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 935             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 936             fvdw12           = _mm_mul_ps(c12_00,FF);
 937             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 938
 939             fscal            = _mm_add_ps(felec,fvdw);
 940
 941             /* Calculate temporary vectorial force */
 942             tx               = _mm_mul_ps(fscal,dx00);
 943             ty               = _mm_mul_ps(fscal,dy00);
 944             tz               = _mm_mul_ps(fscal,dz00);
 945
 946             /* Update vectorial force */
 947             fix0             = _mm_add_ps(fix0,tx);
 948             fiy0             = _mm_add_ps(fiy0,ty);
 949             fiz0             = _mm_add_ps(fiz0,tz);
 950
 951             fjptrA             = f+j_coord_offsetA;
 952             fjptrB             = f+j_coord_offsetB;
 953             fjptrC             = f+j_coord_offsetC;
 954             fjptrD             = f+j_coord_offsetD;
 955             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             r10              = _mm_mul_ps(rsq10,rinv10);
 962
 963             /* Compute parameters for interactions between i and j atoms */
 964             qq10             = _mm_mul_ps(iq1,jq0);
 965
 966             /* EWALD ELECTROSTATICS */
 967
 968             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 969             ewrt             = _mm_mul_ps(r10,ewtabscale);
 970             ewitab           = _mm_cvttps_epi32(ewrt);
 971             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 972             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 973                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 974                                          &ewtabF,&ewtabFn);
 975             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 976             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 977
 978             fscal            = felec;
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm_mul_ps(fscal,dx10);
 982             ty               = _mm_mul_ps(fscal,dy10);
 983             tz               = _mm_mul_ps(fscal,dz10);
 984
 985             /* Update vectorial force */
 986             fix1             = _mm_add_ps(fix1,tx);
 987             fiy1             = _mm_add_ps(fiy1,ty);
 988             fiz1             = _mm_add_ps(fiz1,tz);
 989
 990             fjptrA             = f+j_coord_offsetA;
 991             fjptrB             = f+j_coord_offsetB;
 992             fjptrC             = f+j_coord_offsetC;
 993             fjptrD             = f+j_coord_offsetD;
 994             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 995
 996             /**************************
 997              * CALCULATE INTERACTIONS *
 998              **************************/
 999
1000             r20              = _mm_mul_ps(rsq20,rinv20);
1001
1002             /* Compute parameters for interactions between i and j atoms */
1003             qq20             = _mm_mul_ps(iq2,jq0);
1004
1005             /* EWALD ELECTROSTATICS */
1006
1007             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1008             ewrt             = _mm_mul_ps(r20,ewtabscale);
1009             ewitab           = _mm_cvttps_epi32(ewrt);
1010             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1011             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1012                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1013                                          &ewtabF,&ewtabFn);
1014             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1015             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1016
1017             fscal            = felec;
1018
1019             /* Calculate temporary vectorial force */
1020             tx               = _mm_mul_ps(fscal,dx20);
1021             ty               = _mm_mul_ps(fscal,dy20);
1022             tz               = _mm_mul_ps(fscal,dz20);
1023
1024             /* Update vectorial force */
1025             fix2             = _mm_add_ps(fix2,tx);
1026             fiy2             = _mm_add_ps(fiy2,ty);
1027             fiz2             = _mm_add_ps(fiz2,tz);
1028
1029             fjptrA             = f+j_coord_offsetA;
1030             fjptrB             = f+j_coord_offsetB;
1031             fjptrC             = f+j_coord_offsetC;
1032             fjptrD             = f+j_coord_offsetD;
1033             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1034
1035             /* Inner loop uses 134 flops */
1036         }
1037
1038         if(jidx<j_index_end)
1039         {
1040
1041             /* Get j neighbor index, and coordinate index */
1042             jnrlistA         = jjnr[jidx];
1043             jnrlistB         = jjnr[jidx+1];
1044             jnrlistC         = jjnr[jidx+2];
1045             jnrlistD         = jjnr[jidx+3];
1046             /* Sign of each element will be negative for non-real atoms.
1047              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1048              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1049              */
1050             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1051             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1052             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1053             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1054             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1055             j_coord_offsetA  = DIM*jnrA;
1056             j_coord_offsetB  = DIM*jnrB;
1057             j_coord_offsetC  = DIM*jnrC;
1058             j_coord_offsetD  = DIM*jnrD;
1059
1060             /* load j atom coordinates */
1061             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1062                                               x+j_coord_offsetC,x+j_coord_offsetD,
1063                                               &jx0,&jy0,&jz0);
1064
1065             /* Calculate displacement vector */
1066             dx00             = _mm_sub_ps(ix0,jx0);
1067             dy00             = _mm_sub_ps(iy0,jy0);
1068             dz00             = _mm_sub_ps(iz0,jz0);
1069             dx10             = _mm_sub_ps(ix1,jx0);
1070             dy10             = _mm_sub_ps(iy1,jy0);
1071             dz10             = _mm_sub_ps(iz1,jz0);
1072             dx20             = _mm_sub_ps(ix2,jx0);
1073             dy20             = _mm_sub_ps(iy2,jy0);
1074             dz20             = _mm_sub_ps(iz2,jz0);
1075
1076             /* Calculate squared distance and things based on it */
1077             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1078             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1079             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1080
1081             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1082             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1083             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1084
1085             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1086             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1087             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1088
1089             /* Load parameters for j particles */
1090             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1091                                                               charge+jnrC+0,charge+jnrD+0);
1092             vdwjidx0A        = 2*vdwtype[jnrA+0];
1093             vdwjidx0B        = 2*vdwtype[jnrB+0];
1094             vdwjidx0C        = 2*vdwtype[jnrC+0];
1095             vdwjidx0D        = 2*vdwtype[jnrD+0];
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             r00              = _mm_mul_ps(rsq00,rinv00);
1102             r00              = _mm_andnot_ps(dummy_mask,r00);
1103
1104             /* Compute parameters for interactions between i and j atoms */
1105             qq00             = _mm_mul_ps(iq0,jq0);
1106             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1107                                          vdwparam+vdwioffset0+vdwjidx0B,
1108                                          vdwparam+vdwioffset0+vdwjidx0C,
1109                                          vdwparam+vdwioffset0+vdwjidx0D,
1110                                          &c6_00,&c12_00);
1111
1112             /* Calculate table index by multiplying r with table scale and truncate to integer */
1113             rt               = _mm_mul_ps(r00,vftabscale);
1114             vfitab           = _mm_cvttps_epi32(rt);
1115             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1116             vfitab           = _mm_slli_epi32(vfitab,3);
1117
1118             /* EWALD ELECTROSTATICS */
1119
1120             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1121             ewrt             = _mm_mul_ps(r00,ewtabscale);
1122             ewitab           = _mm_cvttps_epi32(ewrt);
1123             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1124             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1125                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1126                                          &ewtabF,&ewtabFn);
1127             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1128             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1129
1130             /* CUBIC SPLINE TABLE DISPERSION */
1131             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1132             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1133             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1134             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1135             _MM_TRANSPOSE4_PS(Y,F,G,H);
1136             Heps             = _mm_mul_ps(vfeps,H);
1137             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1138             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1139             fvdw6            = _mm_mul_ps(c6_00,FF);
1140
1141             /* CUBIC SPLINE TABLE REPULSION */
1142             vfitab           = _mm_add_epi32(vfitab,ifour);
1143             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1144             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1145             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1146             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1147             _MM_TRANSPOSE4_PS(Y,F,G,H);
1148             Heps             = _mm_mul_ps(vfeps,H);
1149             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1150             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1151             fvdw12           = _mm_mul_ps(c12_00,FF);
1152             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1153
1154             fscal            = _mm_add_ps(felec,fvdw);
1155
1156             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1157
1158             /* Calculate temporary vectorial force */
1159             tx               = _mm_mul_ps(fscal,dx00);
1160             ty               = _mm_mul_ps(fscal,dy00);
1161             tz               = _mm_mul_ps(fscal,dz00);
1162
1163             /* Update vectorial force */
1164             fix0             = _mm_add_ps(fix0,tx);
1165             fiy0             = _mm_add_ps(fiy0,ty);
1166             fiz0             = _mm_add_ps(fiz0,tz);
1167
1168             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1169             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1170             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1171             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1172             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1173
1174             /**************************
1175              * CALCULATE INTERACTIONS *
1176              **************************/
1177
1178             r10              = _mm_mul_ps(rsq10,rinv10);
1179             r10              = _mm_andnot_ps(dummy_mask,r10);
1180
1181             /* Compute parameters for interactions between i and j atoms */
1182             qq10             = _mm_mul_ps(iq1,jq0);
1183
1184             /* EWALD ELECTROSTATICS */
1185
1186             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1187             ewrt             = _mm_mul_ps(r10,ewtabscale);
1188             ewitab           = _mm_cvttps_epi32(ewrt);
1189             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1190             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1191                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1192                                          &ewtabF,&ewtabFn);
1193             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1194             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1195
1196             fscal            = felec;
1197
1198             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1199
1200             /* Calculate temporary vectorial force */
1201             tx               = _mm_mul_ps(fscal,dx10);
1202             ty               = _mm_mul_ps(fscal,dy10);
1203             tz               = _mm_mul_ps(fscal,dz10);
1204
1205             /* Update vectorial force */
1206             fix1             = _mm_add_ps(fix1,tx);
1207             fiy1             = _mm_add_ps(fiy1,ty);
1208             fiz1             = _mm_add_ps(fiz1,tz);
1209
1210             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1211             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1212             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1213             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1214             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1215
1216             /**************************
1217              * CALCULATE INTERACTIONS *
1218              **************************/
1219
1220             r20              = _mm_mul_ps(rsq20,rinv20);
1221             r20              = _mm_andnot_ps(dummy_mask,r20);
1222
1223             /* Compute parameters for interactions between i and j atoms */
1224             qq20             = _mm_mul_ps(iq2,jq0);
1225
1226             /* EWALD ELECTROSTATICS */
1227
1228             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1229             ewrt             = _mm_mul_ps(r20,ewtabscale);
1230             ewitab           = _mm_cvttps_epi32(ewrt);
1231             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1232             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1233                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1234                                          &ewtabF,&ewtabFn);
1235             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1236             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1237
1238             fscal            = felec;
1239
1240             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1241
1242             /* Calculate temporary vectorial force */
1243             tx               = _mm_mul_ps(fscal,dx20);
1244             ty               = _mm_mul_ps(fscal,dy20);
1245             tz               = _mm_mul_ps(fscal,dz20);
1246
1247             /* Update vectorial force */
1248             fix2             = _mm_add_ps(fix2,tx);
1249             fiy2             = _mm_add_ps(fiy2,ty);
1250             fiz2             = _mm_add_ps(fiz2,tz);
1251
1252             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1253             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1254             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1255             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1256             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1257
1258             /* Inner loop uses 137 flops */
1259         }
1260
1261         /* End of innermost loop */
1262
1263         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1264                                               f+i_coord_offset,fshift+i_shift_offset);
1265
1266         /* Increment number of inner iterations */
1267         inneriter                  += j_index_end - j_index_start;
1268
1269         /* Outer loop uses 18 flops */
1270     }
1271
1272     /* Increment number of outer iterations */
1273     outeriter        += nri;
1274
1275     /* Update outer/inner flops */
1276
1277     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1278 }