src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 128     rcutoff_scalar   = fr->rcoulomb;
 129     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 130     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 131
 132     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 133     rvdw             = _mm_set1_ps(fr->rvdw);
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 167
 168         fix0             = _mm_setzero_ps();
 169         fiy0             = _mm_setzero_ps();
 170         fiz0             = _mm_setzero_ps();
 171         fix1             = _mm_setzero_ps();
 172         fiy1             = _mm_setzero_ps();
 173         fiz1             = _mm_setzero_ps();
 174         fix2             = _mm_setzero_ps();
 175         fiy2             = _mm_setzero_ps();
 176         fiz2             = _mm_setzero_ps();
 177         fix3             = _mm_setzero_ps();
 178         fiy3             = _mm_setzero_ps();
 179         fiz3             = _mm_setzero_ps();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm_setzero_ps();
 183         vvdwsum          = _mm_setzero_ps();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               x+j_coord_offsetC,x+j_coord_offsetD,
 202                                               &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm_sub_ps(ix0,jx0);
 206             dy00             = _mm_sub_ps(iy0,jy0);
 207             dz00             = _mm_sub_ps(iz0,jz0);
 208             dx10             = _mm_sub_ps(ix1,jx0);
 209             dy10             = _mm_sub_ps(iy1,jy0);
 210             dz10             = _mm_sub_ps(iz1,jz0);
 211             dx20             = _mm_sub_ps(ix2,jx0);
 212             dy20             = _mm_sub_ps(iy2,jy0);
 213             dz20             = _mm_sub_ps(iz2,jz0);
 214             dx30             = _mm_sub_ps(ix3,jx0);
 215             dy30             = _mm_sub_ps(iy3,jy0);
 216             dz30             = _mm_sub_ps(iz3,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 220             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 221             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 222             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 223
 224             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 226             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 227
 228             rinvsq00         = gmx_mm_inv_ps(rsq00);
 229             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 230             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 231             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 235                                                               charge+jnrC+0,charge+jnrD+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             if (gmx_mm_any_lt(rsq00,rcutoff2))
 246             {
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 250                                          vdwparam+vdwioffset0+vdwjidx0B,
 251                                          vdwparam+vdwioffset0+vdwjidx0C,
 252                                          vdwparam+vdwioffset0+vdwjidx0D,
 253                                          &c6_00,&c12_00);
 254
 255             /* LENNARD-JONES DISPERSION/REPULSION */
 256
 257             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 258             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 259             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 260             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 261                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 262             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 263
 264             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 268             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 269
 270             fscal            = fvdw;
 271
 272             fscal            = _mm_and_ps(fscal,cutoff_mask);
 273
 274             /* Calculate temporary vectorial force */
 275             tx               = _mm_mul_ps(fscal,dx00);
 276             ty               = _mm_mul_ps(fscal,dy00);
 277             tz               = _mm_mul_ps(fscal,dz00);
 278
 279             /* Update vectorial force */
 280             fix0             = _mm_add_ps(fix0,tx);
 281             fiy0             = _mm_add_ps(fiy0,ty);
 282             fiz0             = _mm_add_ps(fiz0,tz);
 283
 284             fjptrA             = f+j_coord_offsetA;
 285             fjptrB             = f+j_coord_offsetB;
 286             fjptrC             = f+j_coord_offsetC;
 287             fjptrD             = f+j_coord_offsetD;
 288             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 289
 290             }
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             if (gmx_mm_any_lt(rsq10,rcutoff2))
 297             {
 298
 299             r10              = _mm_mul_ps(rsq10,rinv10);
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq10             = _mm_mul_ps(iq1,jq0);
 303
 304             /* EWALD ELECTROSTATICS */
 305
 306             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 307             ewrt             = _mm_mul_ps(r10,ewtabscale);
 308             ewitab           = _mm_cvttps_epi32(ewrt);
 309             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 310             ewitab           = _mm_slli_epi32(ewitab,2);
 311             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 312             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 313             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 314             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 315             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 316             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 317             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 318             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 319             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 320
 321             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             velec            = _mm_and_ps(velec,cutoff_mask);
 325             velecsum         = _mm_add_ps(velecsum,velec);
 326
 327             fscal            = felec;
 328
 329             fscal            = _mm_and_ps(fscal,cutoff_mask);
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm_mul_ps(fscal,dx10);
 333             ty               = _mm_mul_ps(fscal,dy10);
 334             tz               = _mm_mul_ps(fscal,dz10);
 335
 336             /* Update vectorial force */
 337             fix1             = _mm_add_ps(fix1,tx);
 338             fiy1             = _mm_add_ps(fiy1,ty);
 339             fiz1             = _mm_add_ps(fiz1,tz);
 340
 341             fjptrA             = f+j_coord_offsetA;
 342             fjptrB             = f+j_coord_offsetB;
 343             fjptrC             = f+j_coord_offsetC;
 344             fjptrD             = f+j_coord_offsetD;
 345             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 346
 347             }
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             if (gmx_mm_any_lt(rsq20,rcutoff2))
 354             {
 355
 356             r20              = _mm_mul_ps(rsq20,rinv20);
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq20             = _mm_mul_ps(iq2,jq0);
 360
 361             /* EWALD ELECTROSTATICS */
 362
 363             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 364             ewrt             = _mm_mul_ps(r20,ewtabscale);
 365             ewitab           = _mm_cvttps_epi32(ewrt);
 366             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 367             ewitab           = _mm_slli_epi32(ewitab,2);
 368             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 369             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 370             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 371             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 372             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 373             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 374             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 375             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 376             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 377
 378             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velec            = _mm_and_ps(velec,cutoff_mask);
 382             velecsum         = _mm_add_ps(velecsum,velec);
 383
 384             fscal            = felec;
 385
 386             fscal            = _mm_and_ps(fscal,cutoff_mask);
 387
 388             /* Calculate temporary vectorial force */
 389             tx               = _mm_mul_ps(fscal,dx20);
 390             ty               = _mm_mul_ps(fscal,dy20);
 391             tz               = _mm_mul_ps(fscal,dz20);
 392
 393             /* Update vectorial force */
 394             fix2             = _mm_add_ps(fix2,tx);
 395             fiy2             = _mm_add_ps(fiy2,ty);
 396             fiz2             = _mm_add_ps(fiz2,tz);
 397
 398             fjptrA             = f+j_coord_offsetA;
 399             fjptrB             = f+j_coord_offsetB;
 400             fjptrC             = f+j_coord_offsetC;
 401             fjptrD             = f+j_coord_offsetD;
 402             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 403
 404             }
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             if (gmx_mm_any_lt(rsq30,rcutoff2))
 411             {
 412
 413             r30              = _mm_mul_ps(rsq30,rinv30);
 414
 415             /* Compute parameters for interactions between i and j atoms */
 416             qq30             = _mm_mul_ps(iq3,jq0);
 417
 418             /* EWALD ELECTROSTATICS */
 419
 420             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 421             ewrt             = _mm_mul_ps(r30,ewtabscale);
 422             ewitab           = _mm_cvttps_epi32(ewrt);
 423             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 424             ewitab           = _mm_slli_epi32(ewitab,2);
 425             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 426             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 427             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 428             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 429             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 430             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 431             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 432             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 433             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 434
 435             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 436
 437             /* Update potential sum for this i atom from the interaction with this j atom. */
 438             velec            = _mm_and_ps(velec,cutoff_mask);
 439             velecsum         = _mm_add_ps(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443             fscal            = _mm_and_ps(fscal,cutoff_mask);
 444
 445             /* Calculate temporary vectorial force */
 446             tx               = _mm_mul_ps(fscal,dx30);
 447             ty               = _mm_mul_ps(fscal,dy30);
 448             tz               = _mm_mul_ps(fscal,dz30);
 449
 450             /* Update vectorial force */
 451             fix3             = _mm_add_ps(fix3,tx);
 452             fiy3             = _mm_add_ps(fiy3,ty);
 453             fiz3             = _mm_add_ps(fiz3,tz);
 454
 455             fjptrA             = f+j_coord_offsetA;
 456             fjptrB             = f+j_coord_offsetB;
 457             fjptrC             = f+j_coord_offsetC;
 458             fjptrD             = f+j_coord_offsetD;
 459             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 460
 461             }
 462
 463             /* Inner loop uses 179 flops */
 464         }
 465
 466         if(jidx<j_index_end)
 467         {
 468
 469             /* Get j neighbor index, and coordinate index */
 470             jnrlistA         = jjnr[jidx];
 471             jnrlistB         = jjnr[jidx+1];
 472             jnrlistC         = jjnr[jidx+2];
 473             jnrlistD         = jjnr[jidx+3];
 474             /* Sign of each element will be negative for non-real atoms.
 475              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 476              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 477              */
 478             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 479             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 480             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 481             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 482             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 483             j_coord_offsetA  = DIM*jnrA;
 484             j_coord_offsetB  = DIM*jnrB;
 485             j_coord_offsetC  = DIM*jnrC;
 486             j_coord_offsetD  = DIM*jnrD;
 487
 488             /* load j atom coordinates */
 489             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 490                                               x+j_coord_offsetC,x+j_coord_offsetD,
 491                                               &jx0,&jy0,&jz0);
 492
 493             /* Calculate displacement vector */
 494             dx00             = _mm_sub_ps(ix0,jx0);
 495             dy00             = _mm_sub_ps(iy0,jy0);
 496             dz00             = _mm_sub_ps(iz0,jz0);
 497             dx10             = _mm_sub_ps(ix1,jx0);
 498             dy10             = _mm_sub_ps(iy1,jy0);
 499             dz10             = _mm_sub_ps(iz1,jz0);
 500             dx20             = _mm_sub_ps(ix2,jx0);
 501             dy20             = _mm_sub_ps(iy2,jy0);
 502             dz20             = _mm_sub_ps(iz2,jz0);
 503             dx30             = _mm_sub_ps(ix3,jx0);
 504             dy30             = _mm_sub_ps(iy3,jy0);
 505             dz30             = _mm_sub_ps(iz3,jz0);
 506
 507             /* Calculate squared distance and things based on it */
 508             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 509             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 510             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 511             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 512
 513             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 514             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 515             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 516
 517             rinvsq00         = gmx_mm_inv_ps(rsq00);
 518             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 519             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 520             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 521
 522             /* Load parameters for j particles */
 523             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 524                                                               charge+jnrC+0,charge+jnrD+0);
 525             vdwjidx0A        = 2*vdwtype[jnrA+0];
 526             vdwjidx0B        = 2*vdwtype[jnrB+0];
 527             vdwjidx0C        = 2*vdwtype[jnrC+0];
 528             vdwjidx0D        = 2*vdwtype[jnrD+0];
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             if (gmx_mm_any_lt(rsq00,rcutoff2))
 535             {
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 539                                          vdwparam+vdwioffset0+vdwjidx0B,
 540                                          vdwparam+vdwioffset0+vdwjidx0C,
 541                                          vdwparam+vdwioffset0+vdwjidx0D,
 542                                          &c6_00,&c12_00);
 543
 544             /* LENNARD-JONES DISPERSION/REPULSION */
 545
 546             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 547             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 548             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 549             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 550                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 551             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 552
 553             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 554
 555             /* Update potential sum for this i atom from the interaction with this j atom. */
 556             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 557             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 558             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 559
 560             fscal            = fvdw;
 561
 562             fscal            = _mm_and_ps(fscal,cutoff_mask);
 563
 564             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm_mul_ps(fscal,dx00);
 568             ty               = _mm_mul_ps(fscal,dy00);
 569             tz               = _mm_mul_ps(fscal,dz00);
 570
 571             /* Update vectorial force */
 572             fix0             = _mm_add_ps(fix0,tx);
 573             fiy0             = _mm_add_ps(fiy0,ty);
 574             fiz0             = _mm_add_ps(fiz0,tz);
 575
 576             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 577             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 578             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 579             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 580             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 581
 582             }
 583
 584             /**************************
 585              * CALCULATE INTERACTIONS *
 586              **************************/
 587
 588             if (gmx_mm_any_lt(rsq10,rcutoff2))
 589             {
 590
 591             r10              = _mm_mul_ps(rsq10,rinv10);
 592             r10              = _mm_andnot_ps(dummy_mask,r10);
 593
 594             /* Compute parameters for interactions between i and j atoms */
 595             qq10             = _mm_mul_ps(iq1,jq0);
 596
 597             /* EWALD ELECTROSTATICS */
 598
 599             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 600             ewrt             = _mm_mul_ps(r10,ewtabscale);
 601             ewitab           = _mm_cvttps_epi32(ewrt);
 602             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 603             ewitab           = _mm_slli_epi32(ewitab,2);
 604             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 605             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 606             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 607             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 608             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 609             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 610             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 611             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 612             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 613
 614             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 615
 616             /* Update potential sum for this i atom from the interaction with this j atom. */
 617             velec            = _mm_and_ps(velec,cutoff_mask);
 618             velec            = _mm_andnot_ps(dummy_mask,velec);
 619             velecsum         = _mm_add_ps(velecsum,velec);
 620
 621             fscal            = felec;
 622
 623             fscal            = _mm_and_ps(fscal,cutoff_mask);
 624
 625             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 626
 627             /* Calculate temporary vectorial force */
 628             tx               = _mm_mul_ps(fscal,dx10);
 629             ty               = _mm_mul_ps(fscal,dy10);
 630             tz               = _mm_mul_ps(fscal,dz10);
 631
 632             /* Update vectorial force */
 633             fix1             = _mm_add_ps(fix1,tx);
 634             fiy1             = _mm_add_ps(fiy1,ty);
 635             fiz1             = _mm_add_ps(fiz1,tz);
 636
 637             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 638             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 639             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 640             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 641             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 642
 643             }
 644
 645             /**************************
 646              * CALCULATE INTERACTIONS *
 647              **************************/
 648
 649             if (gmx_mm_any_lt(rsq20,rcutoff2))
 650             {
 651
 652             r20              = _mm_mul_ps(rsq20,rinv20);
 653             r20              = _mm_andnot_ps(dummy_mask,r20);
 654
 655             /* Compute parameters for interactions between i and j atoms */
 656             qq20             = _mm_mul_ps(iq2,jq0);
 657
 658             /* EWALD ELECTROSTATICS */
 659
 660             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 661             ewrt             = _mm_mul_ps(r20,ewtabscale);
 662             ewitab           = _mm_cvttps_epi32(ewrt);
 663             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 664             ewitab           = _mm_slli_epi32(ewitab,2);
 665             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 666             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 667             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 668             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 669             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 670             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 671             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 672             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 673             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 674
 675             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 676
 677             /* Update potential sum for this i atom from the interaction with this j atom. */
 678             velec            = _mm_and_ps(velec,cutoff_mask);
 679             velec            = _mm_andnot_ps(dummy_mask,velec);
 680             velecsum         = _mm_add_ps(velecsum,velec);
 681
 682             fscal            = felec;
 683
 684             fscal            = _mm_and_ps(fscal,cutoff_mask);
 685
 686             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 687
 688             /* Calculate temporary vectorial force */
 689             tx               = _mm_mul_ps(fscal,dx20);
 690             ty               = _mm_mul_ps(fscal,dy20);
 691             tz               = _mm_mul_ps(fscal,dz20);
 692
 693             /* Update vectorial force */
 694             fix2             = _mm_add_ps(fix2,tx);
 695             fiy2             = _mm_add_ps(fiy2,ty);
 696             fiz2             = _mm_add_ps(fiz2,tz);
 697
 698             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 699             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 700             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 701             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 702             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 703
 704             }
 705
 706             /**************************
 707              * CALCULATE INTERACTIONS *
 708              **************************/
 709
 710             if (gmx_mm_any_lt(rsq30,rcutoff2))
 711             {
 712
 713             r30              = _mm_mul_ps(rsq30,rinv30);
 714             r30              = _mm_andnot_ps(dummy_mask,r30);
 715
 716             /* Compute parameters for interactions between i and j atoms */
 717             qq30             = _mm_mul_ps(iq3,jq0);
 718
 719             /* EWALD ELECTROSTATICS */
 720
 721             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 722             ewrt             = _mm_mul_ps(r30,ewtabscale);
 723             ewitab           = _mm_cvttps_epi32(ewrt);
 724             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 725             ewitab           = _mm_slli_epi32(ewitab,2);
 726             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 727             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 728             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 729             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 730             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 731             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 732             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 733             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 734             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 735
 736             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 737
 738             /* Update potential sum for this i atom from the interaction with this j atom. */
 739             velec            = _mm_and_ps(velec,cutoff_mask);
 740             velec            = _mm_andnot_ps(dummy_mask,velec);
 741             velecsum         = _mm_add_ps(velecsum,velec);
 742
 743             fscal            = felec;
 744
 745             fscal            = _mm_and_ps(fscal,cutoff_mask);
 746
 747             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 748
 749             /* Calculate temporary vectorial force */
 750             tx               = _mm_mul_ps(fscal,dx30);
 751             ty               = _mm_mul_ps(fscal,dy30);
 752             tz               = _mm_mul_ps(fscal,dz30);
 753
 754             /* Update vectorial force */
 755             fix3             = _mm_add_ps(fix3,tx);
 756             fiy3             = _mm_add_ps(fiy3,ty);
 757             fiz3             = _mm_add_ps(fiz3,tz);
 758
 759             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 760             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 761             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 762             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 763             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 764
 765             }
 766
 767             /* Inner loop uses 182 flops */
 768         }
 769
 770         /* End of innermost loop */
 771
 772         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 773                                               f+i_coord_offset,fshift+i_shift_offset);
 774
 775         ggid                        = gid[iidx];
 776         /* Update potential energies */
 777         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 778         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 779
 780         /* Increment number of inner iterations */
 781         inneriter                  += j_index_end - j_index_start;
 782
 783         /* Outer loop uses 26 flops */
 784     }
 785
 786     /* Increment number of outer iterations */
 787     outeriter        += nri;
 788
 789     /* Update outer/inner flops */
 790
 791     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*182);
 792 }
 793 /*
 794  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_single
 795  * Electrostatics interaction: Ewald
 796  * VdW interaction:            LennardJones
 797  * Geometry:                   Water4-Particle
 798  * Calculate force/pot:        Force
 799  */
 800 void
 801 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_single
 802                     (t_nblist * gmx_restrict                nlist,
 803                      rvec * gmx_restrict                    xx,
 804                      rvec * gmx_restrict                    ff,
 805                      t_forcerec * gmx_restrict              fr,
 806                      t_mdatoms * gmx_restrict               mdatoms,
 807                      nb_kernel_data_t * gmx_restrict        kernel_data,
 808                      t_nrnb * gmx_restrict                  nrnb)
 809 {
 810     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 811      * just 0 for non-waters.
 812      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 813      * jnr indices corresponding to data put in the four positions in the SIMD register.
 814      */
 815     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 816     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 817     int              jnrA,jnrB,jnrC,jnrD;
 818     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 819     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 820     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 821     real             rcutoff_scalar;
 822     real             *shiftvec,*fshift,*x,*f;
 823     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 824     real             scratch[4*DIM];
 825     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 826     int              vdwioffset0;
 827     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 828     int              vdwioffset1;
 829     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 830     int              vdwioffset2;
 831     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 832     int              vdwioffset3;
 833     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 834     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 835     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 836     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 837     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 838     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 839     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 840     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 841     real             *charge;
 842     int              nvdwtype;
 843     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 844     int              *vdwtype;
 845     real             *vdwparam;
 846     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 847     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 848     __m128i          ewitab;
 849     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 850     real             *ewtab;
 851     __m128           dummy_mask,cutoff_mask;
 852     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 853     __m128           one     = _mm_set1_ps(1.0);
 854     __m128           two     = _mm_set1_ps(2.0);
 855     x                = xx[0];
 856     f                = ff[0];
 857
 858     nri              = nlist->nri;
 859     iinr             = nlist->iinr;
 860     jindex           = nlist->jindex;
 861     jjnr             = nlist->jjnr;
 862     shiftidx         = nlist->shift;
 863     gid              = nlist->gid;
 864     shiftvec         = fr->shift_vec[0];
 865     fshift           = fr->fshift[0];
 866     facel            = _mm_set1_ps(fr->epsfac);
 867     charge           = mdatoms->chargeA;
 868     nvdwtype         = fr->ntype;
 869     vdwparam         = fr->nbfp;
 870     vdwtype          = mdatoms->typeA;
 871
 872     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 873     ewtab            = fr->ic->tabq_coul_F;
 874     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 875     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 876
 877     /* Setup water-specific parameters */
 878     inr              = nlist->iinr[0];
 879     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 880     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 881     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 882     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 883
 884     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 885     rcutoff_scalar   = fr->rcoulomb;
 886     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 887     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 888
 889     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 890     rvdw             = _mm_set1_ps(fr->rvdw);
 891
 892     /* Avoid stupid compiler warnings */
 893     jnrA = jnrB = jnrC = jnrD = 0;
 894     j_coord_offsetA = 0;
 895     j_coord_offsetB = 0;
 896     j_coord_offsetC = 0;
 897     j_coord_offsetD = 0;
 898
 899     outeriter        = 0;
 900     inneriter        = 0;
 901
 902     for(iidx=0;iidx<4*DIM;iidx++)
 903     {
 904         scratch[iidx] = 0.0;
 905     }
 906
 907     /* Start outer loop over neighborlists */
 908     for(iidx=0; iidx<nri; iidx++)
 909     {
 910         /* Load shift vector for this list */
 911         i_shift_offset   = DIM*shiftidx[iidx];
 912
 913         /* Load limits for loop over neighbors */
 914         j_index_start    = jindex[iidx];
 915         j_index_end      = jindex[iidx+1];
 916
 917         /* Get outer coordinate index */
 918         inr              = iinr[iidx];
 919         i_coord_offset   = DIM*inr;
 920
 921         /* Load i particle coords and add shift vector */
 922         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 923                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 924
 925         fix0             = _mm_setzero_ps();
 926         fiy0             = _mm_setzero_ps();
 927         fiz0             = _mm_setzero_ps();
 928         fix1             = _mm_setzero_ps();
 929         fiy1             = _mm_setzero_ps();
 930         fiz1             = _mm_setzero_ps();
 931         fix2             = _mm_setzero_ps();
 932         fiy2             = _mm_setzero_ps();
 933         fiz2             = _mm_setzero_ps();
 934         fix3             = _mm_setzero_ps();
 935         fiy3             = _mm_setzero_ps();
 936         fiz3             = _mm_setzero_ps();
 937
 938         /* Start inner kernel loop */
 939         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 940         {
 941
 942             /* Get j neighbor index, and coordinate index */
 943             jnrA             = jjnr[jidx];
 944             jnrB             = jjnr[jidx+1];
 945             jnrC             = jjnr[jidx+2];
 946             jnrD             = jjnr[jidx+3];
 947             j_coord_offsetA  = DIM*jnrA;
 948             j_coord_offsetB  = DIM*jnrB;
 949             j_coord_offsetC  = DIM*jnrC;
 950             j_coord_offsetD  = DIM*jnrD;
 951
 952             /* load j atom coordinates */
 953             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 954                                               x+j_coord_offsetC,x+j_coord_offsetD,
 955                                               &jx0,&jy0,&jz0);
 956
 957             /* Calculate displacement vector */
 958             dx00             = _mm_sub_ps(ix0,jx0);
 959             dy00             = _mm_sub_ps(iy0,jy0);
 960             dz00             = _mm_sub_ps(iz0,jz0);
 961             dx10             = _mm_sub_ps(ix1,jx0);
 962             dy10             = _mm_sub_ps(iy1,jy0);
 963             dz10             = _mm_sub_ps(iz1,jz0);
 964             dx20             = _mm_sub_ps(ix2,jx0);
 965             dy20             = _mm_sub_ps(iy2,jy0);
 966             dz20             = _mm_sub_ps(iz2,jz0);
 967             dx30             = _mm_sub_ps(ix3,jx0);
 968             dy30             = _mm_sub_ps(iy3,jy0);
 969             dz30             = _mm_sub_ps(iz3,jz0);
 970
 971             /* Calculate squared distance and things based on it */
 972             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 973             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 974             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 975             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 976
 977             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 978             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 979             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 980
 981             rinvsq00         = gmx_mm_inv_ps(rsq00);
 982             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 983             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 984             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 985
 986             /* Load parameters for j particles */
 987             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 988                                                               charge+jnrC+0,charge+jnrD+0);
 989             vdwjidx0A        = 2*vdwtype[jnrA+0];
 990             vdwjidx0B        = 2*vdwtype[jnrB+0];
 991             vdwjidx0C        = 2*vdwtype[jnrC+0];
 992             vdwjidx0D        = 2*vdwtype[jnrD+0];
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             if (gmx_mm_any_lt(rsq00,rcutoff2))
 999             {
1000
1001             /* Compute parameters for interactions between i and j atoms */
1002             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1003                                          vdwparam+vdwioffset0+vdwjidx0B,
1004                                          vdwparam+vdwioffset0+vdwjidx0C,
1005                                          vdwparam+vdwioffset0+vdwjidx0D,
1006                                          &c6_00,&c12_00);
1007
1008             /* LENNARD-JONES DISPERSION/REPULSION */
1009
1010             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1011             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1012
1013             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1014
1015             fscal            = fvdw;
1016
1017             fscal            = _mm_and_ps(fscal,cutoff_mask);
1018
1019             /* Calculate temporary vectorial force */
1020             tx               = _mm_mul_ps(fscal,dx00);
1021             ty               = _mm_mul_ps(fscal,dy00);
1022             tz               = _mm_mul_ps(fscal,dz00);
1023
1024             /* Update vectorial force */
1025             fix0             = _mm_add_ps(fix0,tx);
1026             fiy0             = _mm_add_ps(fiy0,ty);
1027             fiz0             = _mm_add_ps(fiz0,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             }
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             if (gmx_mm_any_lt(rsq10,rcutoff2))
1042             {
1043
1044             r10              = _mm_mul_ps(rsq10,rinv10);
1045
1046             /* Compute parameters for interactions between i and j atoms */
1047             qq10             = _mm_mul_ps(iq1,jq0);
1048
1049             /* EWALD ELECTROSTATICS */
1050
1051             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1052             ewrt             = _mm_mul_ps(r10,ewtabscale);
1053             ewitab           = _mm_cvttps_epi32(ewrt);
1054             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1055             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1056                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1057                                          &ewtabF,&ewtabFn);
1058             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1059             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1060
1061             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1062
1063             fscal            = felec;
1064
1065             fscal            = _mm_and_ps(fscal,cutoff_mask);
1066
1067             /* Calculate temporary vectorial force */
1068             tx               = _mm_mul_ps(fscal,dx10);
1069             ty               = _mm_mul_ps(fscal,dy10);
1070             tz               = _mm_mul_ps(fscal,dz10);
1071
1072             /* Update vectorial force */
1073             fix1             = _mm_add_ps(fix1,tx);
1074             fiy1             = _mm_add_ps(fiy1,ty);
1075             fiz1             = _mm_add_ps(fiz1,tz);
1076
1077             fjptrA             = f+j_coord_offsetA;
1078             fjptrB             = f+j_coord_offsetB;
1079             fjptrC             = f+j_coord_offsetC;
1080             fjptrD             = f+j_coord_offsetD;
1081             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1082
1083             }
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             if (gmx_mm_any_lt(rsq20,rcutoff2))
1090             {
1091
1092             r20              = _mm_mul_ps(rsq20,rinv20);
1093
1094             /* Compute parameters for interactions between i and j atoms */
1095             qq20             = _mm_mul_ps(iq2,jq0);
1096
1097             /* EWALD ELECTROSTATICS */
1098
1099             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1100             ewrt             = _mm_mul_ps(r20,ewtabscale);
1101             ewitab           = _mm_cvttps_epi32(ewrt);
1102             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1103             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1104                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1105                                          &ewtabF,&ewtabFn);
1106             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1107             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1108
1109             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1110
1111             fscal            = felec;
1112
1113             fscal            = _mm_and_ps(fscal,cutoff_mask);
1114
1115             /* Calculate temporary vectorial force */
1116             tx               = _mm_mul_ps(fscal,dx20);
1117             ty               = _mm_mul_ps(fscal,dy20);
1118             tz               = _mm_mul_ps(fscal,dz20);
1119
1120             /* Update vectorial force */
1121             fix2             = _mm_add_ps(fix2,tx);
1122             fiy2             = _mm_add_ps(fiy2,ty);
1123             fiz2             = _mm_add_ps(fiz2,tz);
1124
1125             fjptrA             = f+j_coord_offsetA;
1126             fjptrB             = f+j_coord_offsetB;
1127             fjptrC             = f+j_coord_offsetC;
1128             fjptrD             = f+j_coord_offsetD;
1129             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1130
1131             }
1132
1133             /**************************
1134              * CALCULATE INTERACTIONS *
1135              **************************/
1136
1137             if (gmx_mm_any_lt(rsq30,rcutoff2))
1138             {
1139
1140             r30              = _mm_mul_ps(rsq30,rinv30);
1141
1142             /* Compute parameters for interactions between i and j atoms */
1143             qq30             = _mm_mul_ps(iq3,jq0);
1144
1145             /* EWALD ELECTROSTATICS */
1146
1147             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1148             ewrt             = _mm_mul_ps(r30,ewtabscale);
1149             ewitab           = _mm_cvttps_epi32(ewrt);
1150             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1151             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1152                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1153                                          &ewtabF,&ewtabFn);
1154             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1155             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1156
1157             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1158
1159             fscal            = felec;
1160
1161             fscal            = _mm_and_ps(fscal,cutoff_mask);
1162
1163             /* Calculate temporary vectorial force */
1164             tx               = _mm_mul_ps(fscal,dx30);
1165             ty               = _mm_mul_ps(fscal,dy30);
1166             tz               = _mm_mul_ps(fscal,dz30);
1167
1168             /* Update vectorial force */
1169             fix3             = _mm_add_ps(fix3,tx);
1170             fiy3             = _mm_add_ps(fiy3,ty);
1171             fiz3             = _mm_add_ps(fiz3,tz);
1172
1173             fjptrA             = f+j_coord_offsetA;
1174             fjptrB             = f+j_coord_offsetB;
1175             fjptrC             = f+j_coord_offsetC;
1176             fjptrD             = f+j_coord_offsetD;
1177             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1178
1179             }
1180
1181             /* Inner loop uses 147 flops */
1182         }
1183
1184         if(jidx<j_index_end)
1185         {
1186
1187             /* Get j neighbor index, and coordinate index */
1188             jnrlistA         = jjnr[jidx];
1189             jnrlistB         = jjnr[jidx+1];
1190             jnrlistC         = jjnr[jidx+2];
1191             jnrlistD         = jjnr[jidx+3];
1192             /* Sign of each element will be negative for non-real atoms.
1193              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1194              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1195              */
1196             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1197             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1198             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1199             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1200             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1201             j_coord_offsetA  = DIM*jnrA;
1202             j_coord_offsetB  = DIM*jnrB;
1203             j_coord_offsetC  = DIM*jnrC;
1204             j_coord_offsetD  = DIM*jnrD;
1205
1206             /* load j atom coordinates */
1207             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1208                                               x+j_coord_offsetC,x+j_coord_offsetD,
1209                                               &jx0,&jy0,&jz0);
1210
1211             /* Calculate displacement vector */
1212             dx00             = _mm_sub_ps(ix0,jx0);
1213             dy00             = _mm_sub_ps(iy0,jy0);
1214             dz00             = _mm_sub_ps(iz0,jz0);
1215             dx10             = _mm_sub_ps(ix1,jx0);
1216             dy10             = _mm_sub_ps(iy1,jy0);
1217             dz10             = _mm_sub_ps(iz1,jz0);
1218             dx20             = _mm_sub_ps(ix2,jx0);
1219             dy20             = _mm_sub_ps(iy2,jy0);
1220             dz20             = _mm_sub_ps(iz2,jz0);
1221             dx30             = _mm_sub_ps(ix3,jx0);
1222             dy30             = _mm_sub_ps(iy3,jy0);
1223             dz30             = _mm_sub_ps(iz3,jz0);
1224
1225             /* Calculate squared distance and things based on it */
1226             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1227             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1228             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1229             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1230
1231             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1232             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1233             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1234
1235             rinvsq00         = gmx_mm_inv_ps(rsq00);
1236             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1237             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1238             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1239
1240             /* Load parameters for j particles */
1241             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1242                                                               charge+jnrC+0,charge+jnrD+0);
1243             vdwjidx0A        = 2*vdwtype[jnrA+0];
1244             vdwjidx0B        = 2*vdwtype[jnrB+0];
1245             vdwjidx0C        = 2*vdwtype[jnrC+0];
1246             vdwjidx0D        = 2*vdwtype[jnrD+0];
1247
1248             /**************************
1249              * CALCULATE INTERACTIONS *
1250              **************************/
1251
1252             if (gmx_mm_any_lt(rsq00,rcutoff2))
1253             {
1254
1255             /* Compute parameters for interactions between i and j atoms */
1256             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1257                                          vdwparam+vdwioffset0+vdwjidx0B,
1258                                          vdwparam+vdwioffset0+vdwjidx0C,
1259                                          vdwparam+vdwioffset0+vdwjidx0D,
1260                                          &c6_00,&c12_00);
1261
1262             /* LENNARD-JONES DISPERSION/REPULSION */
1263
1264             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1265             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1266
1267             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1268
1269             fscal            = fvdw;
1270
1271             fscal            = _mm_and_ps(fscal,cutoff_mask);
1272
1273             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1274
1275             /* Calculate temporary vectorial force */
1276             tx               = _mm_mul_ps(fscal,dx00);
1277             ty               = _mm_mul_ps(fscal,dy00);
1278             tz               = _mm_mul_ps(fscal,dz00);
1279
1280             /* Update vectorial force */
1281             fix0             = _mm_add_ps(fix0,tx);
1282             fiy0             = _mm_add_ps(fiy0,ty);
1283             fiz0             = _mm_add_ps(fiz0,tz);
1284
1285             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1286             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1287             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1288             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1289             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1290
1291             }
1292
1293             /**************************
1294              * CALCULATE INTERACTIONS *
1295              **************************/
1296
1297             if (gmx_mm_any_lt(rsq10,rcutoff2))
1298             {
1299
1300             r10              = _mm_mul_ps(rsq10,rinv10);
1301             r10              = _mm_andnot_ps(dummy_mask,r10);
1302
1303             /* Compute parameters for interactions between i and j atoms */
1304             qq10             = _mm_mul_ps(iq1,jq0);
1305
1306             /* EWALD ELECTROSTATICS */
1307
1308             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1309             ewrt             = _mm_mul_ps(r10,ewtabscale);
1310             ewitab           = _mm_cvttps_epi32(ewrt);
1311             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1312             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1313                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1314                                          &ewtabF,&ewtabFn);
1315             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1316             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1317
1318             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1319
1320             fscal            = felec;
1321
1322             fscal            = _mm_and_ps(fscal,cutoff_mask);
1323
1324             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1325
1326             /* Calculate temporary vectorial force */
1327             tx               = _mm_mul_ps(fscal,dx10);
1328             ty               = _mm_mul_ps(fscal,dy10);
1329             tz               = _mm_mul_ps(fscal,dz10);
1330
1331             /* Update vectorial force */
1332             fix1             = _mm_add_ps(fix1,tx);
1333             fiy1             = _mm_add_ps(fiy1,ty);
1334             fiz1             = _mm_add_ps(fiz1,tz);
1335
1336             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1337             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1338             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1339             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1340             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1341
1342             }
1343
1344             /**************************
1345              * CALCULATE INTERACTIONS *
1346              **************************/
1347
1348             if (gmx_mm_any_lt(rsq20,rcutoff2))
1349             {
1350
1351             r20              = _mm_mul_ps(rsq20,rinv20);
1352             r20              = _mm_andnot_ps(dummy_mask,r20);
1353
1354             /* Compute parameters for interactions between i and j atoms */
1355             qq20             = _mm_mul_ps(iq2,jq0);
1356
1357             /* EWALD ELECTROSTATICS */
1358
1359             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1360             ewrt             = _mm_mul_ps(r20,ewtabscale);
1361             ewitab           = _mm_cvttps_epi32(ewrt);
1362             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1363             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1364                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1365                                          &ewtabF,&ewtabFn);
1366             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1367             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1368
1369             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1370
1371             fscal            = felec;
1372
1373             fscal            = _mm_and_ps(fscal,cutoff_mask);
1374
1375             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1376
1377             /* Calculate temporary vectorial force */
1378             tx               = _mm_mul_ps(fscal,dx20);
1379             ty               = _mm_mul_ps(fscal,dy20);
1380             tz               = _mm_mul_ps(fscal,dz20);
1381
1382             /* Update vectorial force */
1383             fix2             = _mm_add_ps(fix2,tx);
1384             fiy2             = _mm_add_ps(fiy2,ty);
1385             fiz2             = _mm_add_ps(fiz2,tz);
1386
1387             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1388             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1389             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1390             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1391             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1392
1393             }
1394
1395             /**************************
1396              * CALCULATE INTERACTIONS *
1397              **************************/
1398
1399             if (gmx_mm_any_lt(rsq30,rcutoff2))
1400             {
1401
1402             r30              = _mm_mul_ps(rsq30,rinv30);
1403             r30              = _mm_andnot_ps(dummy_mask,r30);
1404
1405             /* Compute parameters for interactions between i and j atoms */
1406             qq30             = _mm_mul_ps(iq3,jq0);
1407
1408             /* EWALD ELECTROSTATICS */
1409
1410             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1411             ewrt             = _mm_mul_ps(r30,ewtabscale);
1412             ewitab           = _mm_cvttps_epi32(ewrt);
1413             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1414             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1415                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1416                                          &ewtabF,&ewtabFn);
1417             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1418             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1419
1420             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1421
1422             fscal            = felec;
1423
1424             fscal            = _mm_and_ps(fscal,cutoff_mask);
1425
1426             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1427
1428             /* Calculate temporary vectorial force */
1429             tx               = _mm_mul_ps(fscal,dx30);
1430             ty               = _mm_mul_ps(fscal,dy30);
1431             tz               = _mm_mul_ps(fscal,dz30);
1432
1433             /* Update vectorial force */
1434             fix3             = _mm_add_ps(fix3,tx);
1435             fiy3             = _mm_add_ps(fiy3,ty);
1436             fiz3             = _mm_add_ps(fiz3,tz);
1437
1438             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1439             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1440             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1441             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1442             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1443
1444             }
1445
1446             /* Inner loop uses 150 flops */
1447         }
1448
1449         /* End of innermost loop */
1450
1451         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1452                                               f+i_coord_offset,fshift+i_shift_offset);
1453
1454         /* Increment number of inner iterations */
1455         inneriter                  += j_index_end - j_index_start;
1456
1457         /* Outer loop uses 24 flops */
1458     }
1459
1460     /* Increment number of outer iterations */
1461     outeriter        += nri;
1462
1463     /* Update outer/inner flops */
1464
1465     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1466 }