src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJ_GeomW4P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse4_1_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse4_1_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          ewitab;
  92     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  93     real             *ewtab;
  94     __m128           dummy_mask,cutoff_mask;
  95     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  96     __m128           one     = _mm_set1_ps(1.0);
  97     __m128           two     = _mm_set1_ps(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm_set1_ps(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 116     ewtab            = fr->ic->tabq_coul_FDV0;
 117     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 118     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 123     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 124     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 125     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     for(iidx=0;iidx<4*DIM;iidx++)
 138     {
 139         scratch[iidx] = 0.0;
 140     }
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 158                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 159
 160         fix0             = _mm_setzero_ps();
 161         fiy0             = _mm_setzero_ps();
 162         fiz0             = _mm_setzero_ps();
 163         fix1             = _mm_setzero_ps();
 164         fiy1             = _mm_setzero_ps();
 165         fiz1             = _mm_setzero_ps();
 166         fix2             = _mm_setzero_ps();
 167         fiy2             = _mm_setzero_ps();
 168         fiz2             = _mm_setzero_ps();
 169         fix3             = _mm_setzero_ps();
 170         fiy3             = _mm_setzero_ps();
 171         fiz3             = _mm_setzero_ps();
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm_setzero_ps();
 175         vvdwsum          = _mm_setzero_ps();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             jnrC             = jjnr[jidx+2];
 185             jnrD             = jjnr[jidx+3];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190
 191             /* load j atom coordinates */
 192             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                               x+j_coord_offsetC,x+j_coord_offsetD,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_ps(ix0,jx0);
 198             dy00             = _mm_sub_ps(iy0,jy0);
 199             dz00             = _mm_sub_ps(iz0,jz0);
 200             dx10             = _mm_sub_ps(ix1,jx0);
 201             dy10             = _mm_sub_ps(iy1,jy0);
 202             dz10             = _mm_sub_ps(iz1,jz0);
 203             dx20             = _mm_sub_ps(ix2,jx0);
 204             dy20             = _mm_sub_ps(iy2,jy0);
 205             dz20             = _mm_sub_ps(iz2,jz0);
 206             dx30             = _mm_sub_ps(ix3,jx0);
 207             dy30             = _mm_sub_ps(iy3,jy0);
 208             dz30             = _mm_sub_ps(iz3,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 212             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 213             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 214             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 215
 216             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 217             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 218             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 219
 220             rinvsq00         = gmx_mm_inv_ps(rsq00);
 221             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 223             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                               charge+jnrC+0,charge+jnrD+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 239                                          vdwparam+vdwioffset0+vdwjidx0B,
 240                                          vdwparam+vdwioffset0+vdwjidx0C,
 241                                          vdwparam+vdwioffset0+vdwjidx0D,
 242                                          &c6_00,&c12_00);
 243
 244             /* LENNARD-JONES DISPERSION/REPULSION */
 245
 246             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 247             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 248             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 249             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 250             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 254
 255             fscal            = fvdw;
 256
 257             /* Calculate temporary vectorial force */
 258             tx               = _mm_mul_ps(fscal,dx00);
 259             ty               = _mm_mul_ps(fscal,dy00);
 260             tz               = _mm_mul_ps(fscal,dz00);
 261
 262             /* Update vectorial force */
 263             fix0             = _mm_add_ps(fix0,tx);
 264             fiy0             = _mm_add_ps(fiy0,ty);
 265             fiz0             = _mm_add_ps(fiz0,tz);
 266
 267             fjptrA             = f+j_coord_offsetA;
 268             fjptrB             = f+j_coord_offsetB;
 269             fjptrC             = f+j_coord_offsetC;
 270             fjptrD             = f+j_coord_offsetD;
 271             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             r10              = _mm_mul_ps(rsq10,rinv10);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq10             = _mm_mul_ps(iq1,jq0);
 281
 282             /* EWALD ELECTROSTATICS */
 283
 284             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 285             ewrt             = _mm_mul_ps(r10,ewtabscale);
 286             ewitab           = _mm_cvttps_epi32(ewrt);
 287             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 288             ewitab           = _mm_slli_epi32(ewitab,2);
 289             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 290             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 291             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 292             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 293             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 294             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 295             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 296             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 297             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velecsum         = _mm_add_ps(velecsum,velec);
 301
 302             fscal            = felec;
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm_mul_ps(fscal,dx10);
 306             ty               = _mm_mul_ps(fscal,dy10);
 307             tz               = _mm_mul_ps(fscal,dz10);
 308
 309             /* Update vectorial force */
 310             fix1             = _mm_add_ps(fix1,tx);
 311             fiy1             = _mm_add_ps(fiy1,ty);
 312             fiz1             = _mm_add_ps(fiz1,tz);
 313
 314             fjptrA             = f+j_coord_offsetA;
 315             fjptrB             = f+j_coord_offsetB;
 316             fjptrC             = f+j_coord_offsetC;
 317             fjptrD             = f+j_coord_offsetD;
 318             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             r20              = _mm_mul_ps(rsq20,rinv20);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq20             = _mm_mul_ps(iq2,jq0);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 332             ewrt             = _mm_mul_ps(r20,ewtabscale);
 333             ewitab           = _mm_cvttps_epi32(ewrt);
 334             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 335             ewitab           = _mm_slli_epi32(ewitab,2);
 336             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 337             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 338             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 339             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 340             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 341             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 342             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 343             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 344             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm_mul_ps(fscal,dx20);
 353             ty               = _mm_mul_ps(fscal,dy20);
 354             tz               = _mm_mul_ps(fscal,dz20);
 355
 356             /* Update vectorial force */
 357             fix2             = _mm_add_ps(fix2,tx);
 358             fiy2             = _mm_add_ps(fiy2,ty);
 359             fiz2             = _mm_add_ps(fiz2,tz);
 360
 361             fjptrA             = f+j_coord_offsetA;
 362             fjptrB             = f+j_coord_offsetB;
 363             fjptrC             = f+j_coord_offsetC;
 364             fjptrD             = f+j_coord_offsetD;
 365             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             r30              = _mm_mul_ps(rsq30,rinv30);
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq30             = _mm_mul_ps(iq3,jq0);
 375
 376             /* EWALD ELECTROSTATICS */
 377
 378             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 379             ewrt             = _mm_mul_ps(r30,ewtabscale);
 380             ewitab           = _mm_cvttps_epi32(ewrt);
 381             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 382             ewitab           = _mm_slli_epi32(ewitab,2);
 383             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 384             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 385             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 386             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 387             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 388             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 389             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 390             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 391             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velecsum         = _mm_add_ps(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm_mul_ps(fscal,dx30);
 400             ty               = _mm_mul_ps(fscal,dy30);
 401             tz               = _mm_mul_ps(fscal,dz30);
 402
 403             /* Update vectorial force */
 404             fix3             = _mm_add_ps(fix3,tx);
 405             fiy3             = _mm_add_ps(fiy3,ty);
 406             fiz3             = _mm_add_ps(fiz3,tz);
 407
 408             fjptrA             = f+j_coord_offsetA;
 409             fjptrB             = f+j_coord_offsetB;
 410             fjptrC             = f+j_coord_offsetC;
 411             fjptrD             = f+j_coord_offsetD;
 412             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 413
 414             /* Inner loop uses 155 flops */
 415         }
 416
 417         if(jidx<j_index_end)
 418         {
 419
 420             /* Get j neighbor index, and coordinate index */
 421             jnrlistA         = jjnr[jidx];
 422             jnrlistB         = jjnr[jidx+1];
 423             jnrlistC         = jjnr[jidx+2];
 424             jnrlistD         = jjnr[jidx+3];
 425             /* Sign of each element will be negative for non-real atoms.
 426              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 427              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 428              */
 429             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 430             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 431             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 432             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 433             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 434             j_coord_offsetA  = DIM*jnrA;
 435             j_coord_offsetB  = DIM*jnrB;
 436             j_coord_offsetC  = DIM*jnrC;
 437             j_coord_offsetD  = DIM*jnrD;
 438
 439             /* load j atom coordinates */
 440             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 441                                               x+j_coord_offsetC,x+j_coord_offsetD,
 442                                               &jx0,&jy0,&jz0);
 443
 444             /* Calculate displacement vector */
 445             dx00             = _mm_sub_ps(ix0,jx0);
 446             dy00             = _mm_sub_ps(iy0,jy0);
 447             dz00             = _mm_sub_ps(iz0,jz0);
 448             dx10             = _mm_sub_ps(ix1,jx0);
 449             dy10             = _mm_sub_ps(iy1,jy0);
 450             dz10             = _mm_sub_ps(iz1,jz0);
 451             dx20             = _mm_sub_ps(ix2,jx0);
 452             dy20             = _mm_sub_ps(iy2,jy0);
 453             dz20             = _mm_sub_ps(iz2,jz0);
 454             dx30             = _mm_sub_ps(ix3,jx0);
 455             dy30             = _mm_sub_ps(iy3,jy0);
 456             dz30             = _mm_sub_ps(iz3,jz0);
 457
 458             /* Calculate squared distance and things based on it */
 459             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 460             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 461             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 462             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 463
 464             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 465             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 466             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 467
 468             rinvsq00         = gmx_mm_inv_ps(rsq00);
 469             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 470             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 471             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 472
 473             /* Load parameters for j particles */
 474             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 475                                                               charge+jnrC+0,charge+jnrD+0);
 476             vdwjidx0A        = 2*vdwtype[jnrA+0];
 477             vdwjidx0B        = 2*vdwtype[jnrB+0];
 478             vdwjidx0C        = 2*vdwtype[jnrC+0];
 479             vdwjidx0D        = 2*vdwtype[jnrD+0];
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 487                                          vdwparam+vdwioffset0+vdwjidx0B,
 488                                          vdwparam+vdwioffset0+vdwjidx0C,
 489                                          vdwparam+vdwioffset0+vdwjidx0D,
 490                                          &c6_00,&c12_00);
 491
 492             /* LENNARD-JONES DISPERSION/REPULSION */
 493
 494             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 495             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 496             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 497             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 498             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 499
 500             /* Update potential sum for this i atom from the interaction with this j atom. */
 501             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 502             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 503
 504             fscal            = fvdw;
 505
 506             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm_mul_ps(fscal,dx00);
 510             ty               = _mm_mul_ps(fscal,dy00);
 511             tz               = _mm_mul_ps(fscal,dz00);
 512
 513             /* Update vectorial force */
 514             fix0             = _mm_add_ps(fix0,tx);
 515             fiy0             = _mm_add_ps(fiy0,ty);
 516             fiz0             = _mm_add_ps(fiz0,tz);
 517
 518             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 519             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 520             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 521             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 522             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r10              = _mm_mul_ps(rsq10,rinv10);
 529             r10              = _mm_andnot_ps(dummy_mask,r10);
 530
 531             /* Compute parameters for interactions between i and j atoms */
 532             qq10             = _mm_mul_ps(iq1,jq0);
 533
 534             /* EWALD ELECTROSTATICS */
 535
 536             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 537             ewrt             = _mm_mul_ps(r10,ewtabscale);
 538             ewitab           = _mm_cvttps_epi32(ewrt);
 539             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 540             ewitab           = _mm_slli_epi32(ewitab,2);
 541             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 542             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 543             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 544             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 545             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 546             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 547             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 548             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 549             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm_andnot_ps(dummy_mask,velec);
 553             velecsum         = _mm_add_ps(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm_mul_ps(fscal,dx10);
 561             ty               = _mm_mul_ps(fscal,dy10);
 562             tz               = _mm_mul_ps(fscal,dz10);
 563
 564             /* Update vectorial force */
 565             fix1             = _mm_add_ps(fix1,tx);
 566             fiy1             = _mm_add_ps(fiy1,ty);
 567             fiz1             = _mm_add_ps(fiz1,tz);
 568
 569             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 570             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 571             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 572             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 573             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             r20              = _mm_mul_ps(rsq20,rinv20);
 580             r20              = _mm_andnot_ps(dummy_mask,r20);
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq20             = _mm_mul_ps(iq2,jq0);
 584
 585             /* EWALD ELECTROSTATICS */
 586
 587             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 588             ewrt             = _mm_mul_ps(r20,ewtabscale);
 589             ewitab           = _mm_cvttps_epi32(ewrt);
 590             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 591             ewitab           = _mm_slli_epi32(ewitab,2);
 592             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 593             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 594             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 595             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 596             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 597             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 598             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 599             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 600             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 601
 602             /* Update potential sum for this i atom from the interaction with this j atom. */
 603             velec            = _mm_andnot_ps(dummy_mask,velec);
 604             velecsum         = _mm_add_ps(velecsum,velec);
 605
 606             fscal            = felec;
 607
 608             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 609
 610             /* Calculate temporary vectorial force */
 611             tx               = _mm_mul_ps(fscal,dx20);
 612             ty               = _mm_mul_ps(fscal,dy20);
 613             tz               = _mm_mul_ps(fscal,dz20);
 614
 615             /* Update vectorial force */
 616             fix2             = _mm_add_ps(fix2,tx);
 617             fiy2             = _mm_add_ps(fiy2,ty);
 618             fiz2             = _mm_add_ps(fiz2,tz);
 619
 620             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 621             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 622             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 623             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 624             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             r30              = _mm_mul_ps(rsq30,rinv30);
 631             r30              = _mm_andnot_ps(dummy_mask,r30);
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq30             = _mm_mul_ps(iq3,jq0);
 635
 636             /* EWALD ELECTROSTATICS */
 637
 638             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 639             ewrt             = _mm_mul_ps(r30,ewtabscale);
 640             ewitab           = _mm_cvttps_epi32(ewrt);
 641             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 642             ewitab           = _mm_slli_epi32(ewitab,2);
 643             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 644             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 645             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 646             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 647             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 648             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 649             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 650             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 651             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 652
 653             /* Update potential sum for this i atom from the interaction with this j atom. */
 654             velec            = _mm_andnot_ps(dummy_mask,velec);
 655             velecsum         = _mm_add_ps(velecsum,velec);
 656
 657             fscal            = felec;
 658
 659             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm_mul_ps(fscal,dx30);
 663             ty               = _mm_mul_ps(fscal,dy30);
 664             tz               = _mm_mul_ps(fscal,dz30);
 665
 666             /* Update vectorial force */
 667             fix3             = _mm_add_ps(fix3,tx);
 668             fiy3             = _mm_add_ps(fiy3,ty);
 669             fiz3             = _mm_add_ps(fiz3,tz);
 670
 671             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 672             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 673             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 674             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 675             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 676
 677             /* Inner loop uses 158 flops */
 678         }
 679
 680         /* End of innermost loop */
 681
 682         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 683                                               f+i_coord_offset,fshift+i_shift_offset);
 684
 685         ggid                        = gid[iidx];
 686         /* Update potential energies */
 687         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 688         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 689
 690         /* Increment number of inner iterations */
 691         inneriter                  += j_index_end - j_index_start;
 692
 693         /* Outer loop uses 26 flops */
 694     }
 695
 696     /* Increment number of outer iterations */
 697     outeriter        += nri;
 698
 699     /* Update outer/inner flops */
 700
 701     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*158);
 702 }
 703 /*
 704  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse4_1_single
 705  * Electrostatics interaction: Ewald
 706  * VdW interaction:            LennardJones
 707  * Geometry:                   Water4-Particle
 708  * Calculate force/pot:        Force
 709  */
 710 void
 711 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse4_1_single
 712                     (t_nblist * gmx_restrict                nlist,
 713                      rvec * gmx_restrict                    xx,
 714                      rvec * gmx_restrict                    ff,
 715                      t_forcerec * gmx_restrict              fr,
 716                      t_mdatoms * gmx_restrict               mdatoms,
 717                      nb_kernel_data_t * gmx_restrict        kernel_data,
 718                      t_nrnb * gmx_restrict                  nrnb)
 719 {
 720     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 721      * just 0 for non-waters.
 722      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 723      * jnr indices corresponding to data put in the four positions in the SIMD register.
 724      */
 725     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 726     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 727     int              jnrA,jnrB,jnrC,jnrD;
 728     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 729     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 730     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 731     real             rcutoff_scalar;
 732     real             *shiftvec,*fshift,*x,*f;
 733     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 734     real             scratch[4*DIM];
 735     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 736     int              vdwioffset0;
 737     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 738     int              vdwioffset1;
 739     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 740     int              vdwioffset2;
 741     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 742     int              vdwioffset3;
 743     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 744     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 745     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 746     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 747     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 748     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 749     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 750     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 751     real             *charge;
 752     int              nvdwtype;
 753     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 754     int              *vdwtype;
 755     real             *vdwparam;
 756     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 757     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 758     __m128i          ewitab;
 759     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 760     real             *ewtab;
 761     __m128           dummy_mask,cutoff_mask;
 762     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 763     __m128           one     = _mm_set1_ps(1.0);
 764     __m128           two     = _mm_set1_ps(2.0);
 765     x                = xx[0];
 766     f                = ff[0];
 767
 768     nri              = nlist->nri;
 769     iinr             = nlist->iinr;
 770     jindex           = nlist->jindex;
 771     jjnr             = nlist->jjnr;
 772     shiftidx         = nlist->shift;
 773     gid              = nlist->gid;
 774     shiftvec         = fr->shift_vec[0];
 775     fshift           = fr->fshift[0];
 776     facel            = _mm_set1_ps(fr->epsfac);
 777     charge           = mdatoms->chargeA;
 778     nvdwtype         = fr->ntype;
 779     vdwparam         = fr->nbfp;
 780     vdwtype          = mdatoms->typeA;
 781
 782     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 783     ewtab            = fr->ic->tabq_coul_F;
 784     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 785     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 786
 787     /* Setup water-specific parameters */
 788     inr              = nlist->iinr[0];
 789     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 790     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 791     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 792     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 793
 794     /* Avoid stupid compiler warnings */
 795     jnrA = jnrB = jnrC = jnrD = 0;
 796     j_coord_offsetA = 0;
 797     j_coord_offsetB = 0;
 798     j_coord_offsetC = 0;
 799     j_coord_offsetD = 0;
 800
 801     outeriter        = 0;
 802     inneriter        = 0;
 803
 804     for(iidx=0;iidx<4*DIM;iidx++)
 805     {
 806         scratch[iidx] = 0.0;
 807     }
 808
 809     /* Start outer loop over neighborlists */
 810     for(iidx=0; iidx<nri; iidx++)
 811     {
 812         /* Load shift vector for this list */
 813         i_shift_offset   = DIM*shiftidx[iidx];
 814
 815         /* Load limits for loop over neighbors */
 816         j_index_start    = jindex[iidx];
 817         j_index_end      = jindex[iidx+1];
 818
 819         /* Get outer coordinate index */
 820         inr              = iinr[iidx];
 821         i_coord_offset   = DIM*inr;
 822
 823         /* Load i particle coords and add shift vector */
 824         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 825                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 826
 827         fix0             = _mm_setzero_ps();
 828         fiy0             = _mm_setzero_ps();
 829         fiz0             = _mm_setzero_ps();
 830         fix1             = _mm_setzero_ps();
 831         fiy1             = _mm_setzero_ps();
 832         fiz1             = _mm_setzero_ps();
 833         fix2             = _mm_setzero_ps();
 834         fiy2             = _mm_setzero_ps();
 835         fiz2             = _mm_setzero_ps();
 836         fix3             = _mm_setzero_ps();
 837         fiy3             = _mm_setzero_ps();
 838         fiz3             = _mm_setzero_ps();
 839
 840         /* Start inner kernel loop */
 841         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 842         {
 843
 844             /* Get j neighbor index, and coordinate index */
 845             jnrA             = jjnr[jidx];
 846             jnrB             = jjnr[jidx+1];
 847             jnrC             = jjnr[jidx+2];
 848             jnrD             = jjnr[jidx+3];
 849             j_coord_offsetA  = DIM*jnrA;
 850             j_coord_offsetB  = DIM*jnrB;
 851             j_coord_offsetC  = DIM*jnrC;
 852             j_coord_offsetD  = DIM*jnrD;
 853
 854             /* load j atom coordinates */
 855             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 856                                               x+j_coord_offsetC,x+j_coord_offsetD,
 857                                               &jx0,&jy0,&jz0);
 858
 859             /* Calculate displacement vector */
 860             dx00             = _mm_sub_ps(ix0,jx0);
 861             dy00             = _mm_sub_ps(iy0,jy0);
 862             dz00             = _mm_sub_ps(iz0,jz0);
 863             dx10             = _mm_sub_ps(ix1,jx0);
 864             dy10             = _mm_sub_ps(iy1,jy0);
 865             dz10             = _mm_sub_ps(iz1,jz0);
 866             dx20             = _mm_sub_ps(ix2,jx0);
 867             dy20             = _mm_sub_ps(iy2,jy0);
 868             dz20             = _mm_sub_ps(iz2,jz0);
 869             dx30             = _mm_sub_ps(ix3,jx0);
 870             dy30             = _mm_sub_ps(iy3,jy0);
 871             dz30             = _mm_sub_ps(iz3,jz0);
 872
 873             /* Calculate squared distance and things based on it */
 874             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 875             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 876             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 877             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 878
 879             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 880             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 881             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 882
 883             rinvsq00         = gmx_mm_inv_ps(rsq00);
 884             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 885             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 886             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 887
 888             /* Load parameters for j particles */
 889             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 890                                                               charge+jnrC+0,charge+jnrD+0);
 891             vdwjidx0A        = 2*vdwtype[jnrA+0];
 892             vdwjidx0B        = 2*vdwtype[jnrB+0];
 893             vdwjidx0C        = 2*vdwtype[jnrC+0];
 894             vdwjidx0D        = 2*vdwtype[jnrD+0];
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             /* Compute parameters for interactions between i and j atoms */
 901             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 902                                          vdwparam+vdwioffset0+vdwjidx0B,
 903                                          vdwparam+vdwioffset0+vdwjidx0C,
 904                                          vdwparam+vdwioffset0+vdwjidx0D,
 905                                          &c6_00,&c12_00);
 906
 907             /* LENNARD-JONES DISPERSION/REPULSION */
 908
 909             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 910             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 911
 912             fscal            = fvdw;
 913
 914             /* Calculate temporary vectorial force */
 915             tx               = _mm_mul_ps(fscal,dx00);
 916             ty               = _mm_mul_ps(fscal,dy00);
 917             tz               = _mm_mul_ps(fscal,dz00);
 918
 919             /* Update vectorial force */
 920             fix0             = _mm_add_ps(fix0,tx);
 921             fiy0             = _mm_add_ps(fiy0,ty);
 922             fiz0             = _mm_add_ps(fiz0,tz);
 923
 924             fjptrA             = f+j_coord_offsetA;
 925             fjptrB             = f+j_coord_offsetB;
 926             fjptrC             = f+j_coord_offsetC;
 927             fjptrD             = f+j_coord_offsetD;
 928             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 929
 930             /**************************
 931              * CALCULATE INTERACTIONS *
 932              **************************/
 933
 934             r10              = _mm_mul_ps(rsq10,rinv10);
 935
 936             /* Compute parameters for interactions between i and j atoms */
 937             qq10             = _mm_mul_ps(iq1,jq0);
 938
 939             /* EWALD ELECTROSTATICS */
 940
 941             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 942             ewrt             = _mm_mul_ps(r10,ewtabscale);
 943             ewitab           = _mm_cvttps_epi32(ewrt);
 944             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 945             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 946                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 947                                          &ewtabF,&ewtabFn);
 948             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 949             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 950
 951             fscal            = felec;
 952
 953             /* Calculate temporary vectorial force */
 954             tx               = _mm_mul_ps(fscal,dx10);
 955             ty               = _mm_mul_ps(fscal,dy10);
 956             tz               = _mm_mul_ps(fscal,dz10);
 957
 958             /* Update vectorial force */
 959             fix1             = _mm_add_ps(fix1,tx);
 960             fiy1             = _mm_add_ps(fiy1,ty);
 961             fiz1             = _mm_add_ps(fiz1,tz);
 962
 963             fjptrA             = f+j_coord_offsetA;
 964             fjptrB             = f+j_coord_offsetB;
 965             fjptrC             = f+j_coord_offsetC;
 966             fjptrD             = f+j_coord_offsetD;
 967             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             r20              = _mm_mul_ps(rsq20,rinv20);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq20             = _mm_mul_ps(iq2,jq0);
 977
 978             /* EWALD ELECTROSTATICS */
 979
 980             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 981             ewrt             = _mm_mul_ps(r20,ewtabscale);
 982             ewitab           = _mm_cvttps_epi32(ewrt);
 983             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 984             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 985                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 986                                          &ewtabF,&ewtabFn);
 987             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 988             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 989
 990             fscal            = felec;
 991
 992             /* Calculate temporary vectorial force */
 993             tx               = _mm_mul_ps(fscal,dx20);
 994             ty               = _mm_mul_ps(fscal,dy20);
 995             tz               = _mm_mul_ps(fscal,dz20);
 996
 997             /* Update vectorial force */
 998             fix2             = _mm_add_ps(fix2,tx);
 999             fiy2             = _mm_add_ps(fiy2,ty);
1000             fiz2             = _mm_add_ps(fiz2,tz);
1001
1002             fjptrA             = f+j_coord_offsetA;
1003             fjptrB             = f+j_coord_offsetB;
1004             fjptrC             = f+j_coord_offsetC;
1005             fjptrD             = f+j_coord_offsetD;
1006             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             r30              = _mm_mul_ps(rsq30,rinv30);
1013
1014             /* Compute parameters for interactions between i and j atoms */
1015             qq30             = _mm_mul_ps(iq3,jq0);
1016
1017             /* EWALD ELECTROSTATICS */
1018
1019             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1020             ewrt             = _mm_mul_ps(r30,ewtabscale);
1021             ewitab           = _mm_cvttps_epi32(ewrt);
1022             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1023             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1024                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1025                                          &ewtabF,&ewtabFn);
1026             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1027             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1028
1029             fscal            = felec;
1030
1031             /* Calculate temporary vectorial force */
1032             tx               = _mm_mul_ps(fscal,dx30);
1033             ty               = _mm_mul_ps(fscal,dy30);
1034             tz               = _mm_mul_ps(fscal,dz30);
1035
1036             /* Update vectorial force */
1037             fix3             = _mm_add_ps(fix3,tx);
1038             fiy3             = _mm_add_ps(fiy3,ty);
1039             fiz3             = _mm_add_ps(fiz3,tz);
1040
1041             fjptrA             = f+j_coord_offsetA;
1042             fjptrB             = f+j_coord_offsetB;
1043             fjptrC             = f+j_coord_offsetC;
1044             fjptrD             = f+j_coord_offsetD;
1045             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1046
1047             /* Inner loop uses 135 flops */
1048         }
1049
1050         if(jidx<j_index_end)
1051         {
1052
1053             /* Get j neighbor index, and coordinate index */
1054             jnrlistA         = jjnr[jidx];
1055             jnrlistB         = jjnr[jidx+1];
1056             jnrlistC         = jjnr[jidx+2];
1057             jnrlistD         = jjnr[jidx+3];
1058             /* Sign of each element will be negative for non-real atoms.
1059              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1060              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1061              */
1062             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1063             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1064             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1065             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1066             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1067             j_coord_offsetA  = DIM*jnrA;
1068             j_coord_offsetB  = DIM*jnrB;
1069             j_coord_offsetC  = DIM*jnrC;
1070             j_coord_offsetD  = DIM*jnrD;
1071
1072             /* load j atom coordinates */
1073             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1074                                               x+j_coord_offsetC,x+j_coord_offsetD,
1075                                               &jx0,&jy0,&jz0);
1076
1077             /* Calculate displacement vector */
1078             dx00             = _mm_sub_ps(ix0,jx0);
1079             dy00             = _mm_sub_ps(iy0,jy0);
1080             dz00             = _mm_sub_ps(iz0,jz0);
1081             dx10             = _mm_sub_ps(ix1,jx0);
1082             dy10             = _mm_sub_ps(iy1,jy0);
1083             dz10             = _mm_sub_ps(iz1,jz0);
1084             dx20             = _mm_sub_ps(ix2,jx0);
1085             dy20             = _mm_sub_ps(iy2,jy0);
1086             dz20             = _mm_sub_ps(iz2,jz0);
1087             dx30             = _mm_sub_ps(ix3,jx0);
1088             dy30             = _mm_sub_ps(iy3,jy0);
1089             dz30             = _mm_sub_ps(iz3,jz0);
1090
1091             /* Calculate squared distance and things based on it */
1092             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1093             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1094             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1095             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1096
1097             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1098             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1099             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1100
1101             rinvsq00         = gmx_mm_inv_ps(rsq00);
1102             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1103             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1104             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1105
1106             /* Load parameters for j particles */
1107             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1108                                                               charge+jnrC+0,charge+jnrD+0);
1109             vdwjidx0A        = 2*vdwtype[jnrA+0];
1110             vdwjidx0B        = 2*vdwtype[jnrB+0];
1111             vdwjidx0C        = 2*vdwtype[jnrC+0];
1112             vdwjidx0D        = 2*vdwtype[jnrD+0];
1113
1114             /**************************
1115              * CALCULATE INTERACTIONS *
1116              **************************/
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1120                                          vdwparam+vdwioffset0+vdwjidx0B,
1121                                          vdwparam+vdwioffset0+vdwjidx0C,
1122                                          vdwparam+vdwioffset0+vdwjidx0D,
1123                                          &c6_00,&c12_00);
1124
1125             /* LENNARD-JONES DISPERSION/REPULSION */
1126
1127             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1128             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1129
1130             fscal            = fvdw;
1131
1132             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1133
1134             /* Calculate temporary vectorial force */
1135             tx               = _mm_mul_ps(fscal,dx00);
1136             ty               = _mm_mul_ps(fscal,dy00);
1137             tz               = _mm_mul_ps(fscal,dz00);
1138
1139             /* Update vectorial force */
1140             fix0             = _mm_add_ps(fix0,tx);
1141             fiy0             = _mm_add_ps(fiy0,ty);
1142             fiz0             = _mm_add_ps(fiz0,tz);
1143
1144             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1145             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1146             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1147             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1148             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1149
1150             /**************************
1151              * CALCULATE INTERACTIONS *
1152              **************************/
1153
1154             r10              = _mm_mul_ps(rsq10,rinv10);
1155             r10              = _mm_andnot_ps(dummy_mask,r10);
1156
1157             /* Compute parameters for interactions between i and j atoms */
1158             qq10             = _mm_mul_ps(iq1,jq0);
1159
1160             /* EWALD ELECTROSTATICS */
1161
1162             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1163             ewrt             = _mm_mul_ps(r10,ewtabscale);
1164             ewitab           = _mm_cvttps_epi32(ewrt);
1165             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1166             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1167                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1168                                          &ewtabF,&ewtabFn);
1169             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1170             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1171
1172             fscal            = felec;
1173
1174             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1175
1176             /* Calculate temporary vectorial force */
1177             tx               = _mm_mul_ps(fscal,dx10);
1178             ty               = _mm_mul_ps(fscal,dy10);
1179             tz               = _mm_mul_ps(fscal,dz10);
1180
1181             /* Update vectorial force */
1182             fix1             = _mm_add_ps(fix1,tx);
1183             fiy1             = _mm_add_ps(fiy1,ty);
1184             fiz1             = _mm_add_ps(fiz1,tz);
1185
1186             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1187             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1188             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1189             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1190             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1191
1192             /**************************
1193              * CALCULATE INTERACTIONS *
1194              **************************/
1195
1196             r20              = _mm_mul_ps(rsq20,rinv20);
1197             r20              = _mm_andnot_ps(dummy_mask,r20);
1198
1199             /* Compute parameters for interactions between i and j atoms */
1200             qq20             = _mm_mul_ps(iq2,jq0);
1201
1202             /* EWALD ELECTROSTATICS */
1203
1204             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1205             ewrt             = _mm_mul_ps(r20,ewtabscale);
1206             ewitab           = _mm_cvttps_epi32(ewrt);
1207             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1208             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1209                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1210                                          &ewtabF,&ewtabFn);
1211             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1212             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1213
1214             fscal            = felec;
1215
1216             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1217
1218             /* Calculate temporary vectorial force */
1219             tx               = _mm_mul_ps(fscal,dx20);
1220             ty               = _mm_mul_ps(fscal,dy20);
1221             tz               = _mm_mul_ps(fscal,dz20);
1222
1223             /* Update vectorial force */
1224             fix2             = _mm_add_ps(fix2,tx);
1225             fiy2             = _mm_add_ps(fiy2,ty);
1226             fiz2             = _mm_add_ps(fiz2,tz);
1227
1228             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1229             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1230             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1231             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1232             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1233
1234             /**************************
1235              * CALCULATE INTERACTIONS *
1236              **************************/
1237
1238             r30              = _mm_mul_ps(rsq30,rinv30);
1239             r30              = _mm_andnot_ps(dummy_mask,r30);
1240
1241             /* Compute parameters for interactions between i and j atoms */
1242             qq30             = _mm_mul_ps(iq3,jq0);
1243
1244             /* EWALD ELECTROSTATICS */
1245
1246             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1247             ewrt             = _mm_mul_ps(r30,ewtabscale);
1248             ewitab           = _mm_cvttps_epi32(ewrt);
1249             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1250             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1251                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1252                                          &ewtabF,&ewtabFn);
1253             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1254             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1255
1256             fscal            = felec;
1257
1258             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1259
1260             /* Calculate temporary vectorial force */
1261             tx               = _mm_mul_ps(fscal,dx30);
1262             ty               = _mm_mul_ps(fscal,dy30);
1263             tz               = _mm_mul_ps(fscal,dz30);
1264
1265             /* Update vectorial force */
1266             fix3             = _mm_add_ps(fix3,tx);
1267             fiy3             = _mm_add_ps(fiy3,ty);
1268             fiz3             = _mm_add_ps(fiz3,tz);
1269
1270             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1271             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1272             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1273             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1274             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1275
1276             /* Inner loop uses 138 flops */
1277         }
1278
1279         /* End of innermost loop */
1280
1281         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1282                                               f+i_coord_offset,fshift+i_shift_offset);
1283
1284         /* Increment number of inner iterations */
1285         inneriter                  += j_index_end - j_index_start;
1286
1287         /* Outer loop uses 24 flops */
1288     }
1289
1290     /* Increment number of outer iterations */
1291     outeriter        += nri;
1292
1293     /* Update outer/inner flops */
1294
1295     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*138);
1296 }