src/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             fjx0             = _mm_setzero_ps();
 242             fjy0             = _mm_setzero_ps();
 243             fjz0             = _mm_setzero_ps();
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             if (gmx_mm_any_lt(rsq00,rcutoff2))
 250             {
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 254                                          vdwparam+vdwioffset0+vdwjidx0B,
 255                                          vdwparam+vdwioffset0+vdwjidx0C,
 256                                          vdwparam+vdwioffset0+vdwjidx0D,
 257                                          &c6_00,&c12_00);
 258
 259             /* LENNARD-JONES DISPERSION/REPULSION */
 260
 261             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 262             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 263             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 264             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) ,
 265                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 266             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 267
 268             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 269
 270             /* Update potential sum for this i atom from the interaction with this j atom. */
 271             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 272             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 273
 274             fscal            = fvdw;
 275
 276             fscal            = _mm_and_ps(fscal,cutoff_mask);
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm_mul_ps(fscal,dx00);
 280             ty               = _mm_mul_ps(fscal,dy00);
 281             tz               = _mm_mul_ps(fscal,dz00);
 282
 283             /* Update vectorial force */
 284             fix0             = _mm_add_ps(fix0,tx);
 285             fiy0             = _mm_add_ps(fiy0,ty);
 286             fiz0             = _mm_add_ps(fiz0,tz);
 287
 288             fjx0             = _mm_add_ps(fjx0,tx);
 289             fjy0             = _mm_add_ps(fjy0,ty);
 290             fjz0             = _mm_add_ps(fjz0,tz);
 291
 292             }
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm_any_lt(rsq10,rcutoff2))
 299             {
 300
 301             r10              = _mm_mul_ps(rsq10,rinv10);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm_mul_ps(iq1,jq0);
 305
 306             /* EWALD ELECTROSTATICS */
 307
 308             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 309             ewrt             = _mm_mul_ps(r10,ewtabscale);
 310             ewitab           = _mm_cvttps_epi32(ewrt);
 311             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 312             ewitab           = _mm_slli_epi32(ewitab,2);
 313             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 314             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 315             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 316             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 317             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 318             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 319             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 320             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 321             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 322
 323             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velec            = _mm_and_ps(velec,cutoff_mask);
 327             velecsum         = _mm_add_ps(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             fscal            = _mm_and_ps(fscal,cutoff_mask);
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_ps(fscal,dx10);
 335             ty               = _mm_mul_ps(fscal,dy10);
 336             tz               = _mm_mul_ps(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm_add_ps(fix1,tx);
 340             fiy1             = _mm_add_ps(fiy1,ty);
 341             fiz1             = _mm_add_ps(fiz1,tz);
 342
 343             fjx0             = _mm_add_ps(fjx0,tx);
 344             fjy0             = _mm_add_ps(fjy0,ty);
 345             fjz0             = _mm_add_ps(fjz0,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             fjx0             = _mm_add_ps(fjx0,tx);
 399             fjy0             = _mm_add_ps(fjy0,ty);
 400             fjz0             = _mm_add_ps(fjz0,tz);
 401
 402             }
 403
 404             /**************************
 405              * CALCULATE INTERACTIONS *
 406              **************************/
 407
 408             if (gmx_mm_any_lt(rsq30,rcutoff2))
 409             {
 410
 411             r30              = _mm_mul_ps(rsq30,rinv30);
 412
 413             /* Compute parameters for interactions between i and j atoms */
 414             qq30             = _mm_mul_ps(iq3,jq0);
 415
 416             /* EWALD ELECTROSTATICS */
 417
 418             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 419             ewrt             = _mm_mul_ps(r30,ewtabscale);
 420             ewitab           = _mm_cvttps_epi32(ewrt);
 421             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 422             ewitab           = _mm_slli_epi32(ewitab,2);
 423             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 424             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 425             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 426             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 427             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 428             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 429             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 430             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 431             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 432
 433             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 434
 435             /* Update potential sum for this i atom from the interaction with this j atom. */
 436             velec            = _mm_and_ps(velec,cutoff_mask);
 437             velecsum         = _mm_add_ps(velecsum,velec);
 438
 439             fscal            = felec;
 440
 441             fscal            = _mm_and_ps(fscal,cutoff_mask);
 442
 443             /* Calculate temporary vectorial force */
 444             tx               = _mm_mul_ps(fscal,dx30);
 445             ty               = _mm_mul_ps(fscal,dy30);
 446             tz               = _mm_mul_ps(fscal,dz30);
 447
 448             /* Update vectorial force */
 449             fix3             = _mm_add_ps(fix3,tx);
 450             fiy3             = _mm_add_ps(fiy3,ty);
 451             fiz3             = _mm_add_ps(fiz3,tz);
 452
 453             fjx0             = _mm_add_ps(fjx0,tx);
 454             fjy0             = _mm_add_ps(fjy0,ty);
 455             fjz0             = _mm_add_ps(fjz0,tz);
 456
 457             }
 458
 459             fjptrA             = f+j_coord_offsetA;
 460             fjptrB             = f+j_coord_offsetB;
 461             fjptrC             = f+j_coord_offsetC;
 462             fjptrD             = f+j_coord_offsetD;
 463
 464             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 465
 466             /* Inner loop uses 179 flops */
 467         }
 468
 469         if(jidx<j_index_end)
 470         {
 471
 472             /* Get j neighbor index, and coordinate index */
 473             jnrlistA         = jjnr[jidx];
 474             jnrlistB         = jjnr[jidx+1];
 475             jnrlistC         = jjnr[jidx+2];
 476             jnrlistD         = jjnr[jidx+3];
 477             /* Sign of each element will be negative for non-real atoms.
 478              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 479              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 480              */
 481             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 482             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 483             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 484             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 485             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 486             j_coord_offsetA  = DIM*jnrA;
 487             j_coord_offsetB  = DIM*jnrB;
 488             j_coord_offsetC  = DIM*jnrC;
 489             j_coord_offsetD  = DIM*jnrD;
 490
 491             /* load j atom coordinates */
 492             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 493                                               x+j_coord_offsetC,x+j_coord_offsetD,
 494                                               &jx0,&jy0,&jz0);
 495
 496             /* Calculate displacement vector */
 497             dx00             = _mm_sub_ps(ix0,jx0);
 498             dy00             = _mm_sub_ps(iy0,jy0);
 499             dz00             = _mm_sub_ps(iz0,jz0);
 500             dx10             = _mm_sub_ps(ix1,jx0);
 501             dy10             = _mm_sub_ps(iy1,jy0);
 502             dz10             = _mm_sub_ps(iz1,jz0);
 503             dx20             = _mm_sub_ps(ix2,jx0);
 504             dy20             = _mm_sub_ps(iy2,jy0);
 505             dz20             = _mm_sub_ps(iz2,jz0);
 506             dx30             = _mm_sub_ps(ix3,jx0);
 507             dy30             = _mm_sub_ps(iy3,jy0);
 508             dz30             = _mm_sub_ps(iz3,jz0);
 509
 510             /* Calculate squared distance and things based on it */
 511             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 512             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 513             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 514             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 515
 516             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 517             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 518             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 519
 520             rinvsq00         = gmx_mm_inv_ps(rsq00);
 521             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 522             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 523             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 524
 525             /* Load parameters for j particles */
 526             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 527                                                               charge+jnrC+0,charge+jnrD+0);
 528             vdwjidx0A        = 2*vdwtype[jnrA+0];
 529             vdwjidx0B        = 2*vdwtype[jnrB+0];
 530             vdwjidx0C        = 2*vdwtype[jnrC+0];
 531             vdwjidx0D        = 2*vdwtype[jnrD+0];
 532
 533             fjx0             = _mm_setzero_ps();
 534             fjy0             = _mm_setzero_ps();
 535             fjz0             = _mm_setzero_ps();
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             if (gmx_mm_any_lt(rsq00,rcutoff2))
 542             {
 543
 544             /* Compute parameters for interactions between i and j atoms */
 545             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 546                                          vdwparam+vdwioffset0+vdwjidx0B,
 547                                          vdwparam+vdwioffset0+vdwjidx0C,
 548                                          vdwparam+vdwioffset0+vdwjidx0D,
 549                                          &c6_00,&c12_00);
 550
 551             /* LENNARD-JONES DISPERSION/REPULSION */
 552
 553             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 554             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 555             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 556             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) ,
 557                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 558             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 559
 560             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 564             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 565             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 566
 567             fscal            = fvdw;
 568
 569             fscal            = _mm_and_ps(fscal,cutoff_mask);
 570
 571             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm_mul_ps(fscal,dx00);
 575             ty               = _mm_mul_ps(fscal,dy00);
 576             tz               = _mm_mul_ps(fscal,dz00);
 577
 578             /* Update vectorial force */
 579             fix0             = _mm_add_ps(fix0,tx);
 580             fiy0             = _mm_add_ps(fiy0,ty);
 581             fiz0             = _mm_add_ps(fiz0,tz);
 582
 583             fjx0             = _mm_add_ps(fjx0,tx);
 584             fjy0             = _mm_add_ps(fjy0,ty);
 585             fjz0             = _mm_add_ps(fjz0,tz);
 586
 587             }
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             if (gmx_mm_any_lt(rsq10,rcutoff2))
 594             {
 595
 596             r10              = _mm_mul_ps(rsq10,rinv10);
 597             r10              = _mm_andnot_ps(dummy_mask,r10);
 598
 599             /* Compute parameters for interactions between i and j atoms */
 600             qq10             = _mm_mul_ps(iq1,jq0);
 601
 602             /* EWALD ELECTROSTATICS */
 603
 604             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 605             ewrt             = _mm_mul_ps(r10,ewtabscale);
 606             ewitab           = _mm_cvttps_epi32(ewrt);
 607             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 608             ewitab           = _mm_slli_epi32(ewitab,2);
 609             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 610             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 611             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 612             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 613             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 614             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 615             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 616             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 617             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 618
 619             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             velec            = _mm_and_ps(velec,cutoff_mask);
 623             velec            = _mm_andnot_ps(dummy_mask,velec);
 624             velecsum         = _mm_add_ps(velecsum,velec);
 625
 626             fscal            = felec;
 627
 628             fscal            = _mm_and_ps(fscal,cutoff_mask);
 629
 630             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm_mul_ps(fscal,dx10);
 634             ty               = _mm_mul_ps(fscal,dy10);
 635             tz               = _mm_mul_ps(fscal,dz10);
 636
 637             /* Update vectorial force */
 638             fix1             = _mm_add_ps(fix1,tx);
 639             fiy1             = _mm_add_ps(fiy1,ty);
 640             fiz1             = _mm_add_ps(fiz1,tz);
 641
 642             fjx0             = _mm_add_ps(fjx0,tx);
 643             fjy0             = _mm_add_ps(fjy0,ty);
 644             fjz0             = _mm_add_ps(fjz0,tz);
 645
 646             }
 647
 648             /**************************
 649              * CALCULATE INTERACTIONS *
 650              **************************/
 651
 652             if (gmx_mm_any_lt(rsq20,rcutoff2))
 653             {
 654
 655             r20              = _mm_mul_ps(rsq20,rinv20);
 656             r20              = _mm_andnot_ps(dummy_mask,r20);
 657
 658             /* Compute parameters for interactions between i and j atoms */
 659             qq20             = _mm_mul_ps(iq2,jq0);
 660
 661             /* EWALD ELECTROSTATICS */
 662
 663             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 664             ewrt             = _mm_mul_ps(r20,ewtabscale);
 665             ewitab           = _mm_cvttps_epi32(ewrt);
 666             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 667             ewitab           = _mm_slli_epi32(ewitab,2);
 668             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 669             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 670             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 671             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 672             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 673             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 674             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 675             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 676             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 677
 678             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 679
 680             /* Update potential sum for this i atom from the interaction with this j atom. */
 681             velec            = _mm_and_ps(velec,cutoff_mask);
 682             velec            = _mm_andnot_ps(dummy_mask,velec);
 683             velecsum         = _mm_add_ps(velecsum,velec);
 684
 685             fscal            = felec;
 686
 687             fscal            = _mm_and_ps(fscal,cutoff_mask);
 688
 689             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 690
 691             /* Calculate temporary vectorial force */
 692             tx               = _mm_mul_ps(fscal,dx20);
 693             ty               = _mm_mul_ps(fscal,dy20);
 694             tz               = _mm_mul_ps(fscal,dz20);
 695
 696             /* Update vectorial force */
 697             fix2             = _mm_add_ps(fix2,tx);
 698             fiy2             = _mm_add_ps(fiy2,ty);
 699             fiz2             = _mm_add_ps(fiz2,tz);
 700
 701             fjx0             = _mm_add_ps(fjx0,tx);
 702             fjy0             = _mm_add_ps(fjy0,ty);
 703             fjz0             = _mm_add_ps(fjz0,tz);
 704
 705             }
 706
 707             /**************************
 708              * CALCULATE INTERACTIONS *
 709              **************************/
 710
 711             if (gmx_mm_any_lt(rsq30,rcutoff2))
 712             {
 713
 714             r30              = _mm_mul_ps(rsq30,rinv30);
 715             r30              = _mm_andnot_ps(dummy_mask,r30);
 716
 717             /* Compute parameters for interactions between i and j atoms */
 718             qq30             = _mm_mul_ps(iq3,jq0);
 719
 720             /* EWALD ELECTROSTATICS */
 721
 722             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 723             ewrt             = _mm_mul_ps(r30,ewtabscale);
 724             ewitab           = _mm_cvttps_epi32(ewrt);
 725             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 726             ewitab           = _mm_slli_epi32(ewitab,2);
 727             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 728             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 729             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 730             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 731             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 732             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 733             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 734             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 735             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 736
 737             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 738
 739             /* Update potential sum for this i atom from the interaction with this j atom. */
 740             velec            = _mm_and_ps(velec,cutoff_mask);
 741             velec            = _mm_andnot_ps(dummy_mask,velec);
 742             velecsum         = _mm_add_ps(velecsum,velec);
 743
 744             fscal            = felec;
 745
 746             fscal            = _mm_and_ps(fscal,cutoff_mask);
 747
 748             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 749
 750             /* Calculate temporary vectorial force */
 751             tx               = _mm_mul_ps(fscal,dx30);
 752             ty               = _mm_mul_ps(fscal,dy30);
 753             tz               = _mm_mul_ps(fscal,dz30);
 754
 755             /* Update vectorial force */
 756             fix3             = _mm_add_ps(fix3,tx);
 757             fiy3             = _mm_add_ps(fiy3,ty);
 758             fiz3             = _mm_add_ps(fiz3,tz);
 759
 760             fjx0             = _mm_add_ps(fjx0,tx);
 761             fjy0             = _mm_add_ps(fjy0,ty);
 762             fjz0             = _mm_add_ps(fjz0,tz);
 763
 764             }
 765
 766             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 767             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 768             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 769             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 770
 771             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 772
 773             /* Inner loop uses 182 flops */
 774         }
 775
 776         /* End of innermost loop */
 777
 778         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 779                                               f+i_coord_offset,fshift+i_shift_offset);
 780
 781         ggid                        = gid[iidx];
 782         /* Update potential energies */
 783         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 784         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 785
 786         /* Increment number of inner iterations */
 787         inneriter                  += j_index_end - j_index_start;
 788
 789         /* Outer loop uses 26 flops */
 790     }
 791
 792     /* Increment number of outer iterations */
 793     outeriter        += nri;
 794
 795     /* Update outer/inner flops */
 796
 797     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*182);
 798 }
 799 /*
 800  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_single
 801  * Electrostatics interaction: Ewald
 802  * VdW interaction:            LennardJones
 803  * Geometry:                   Water4-Particle
 804  * Calculate force/pot:        Force
 805  */
 806 void
 807 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_sse2_single
 808                     (t_nblist * gmx_restrict                nlist,
 809                      rvec * gmx_restrict                    xx,
 810                      rvec * gmx_restrict                    ff,
 811                      t_forcerec * gmx_restrict              fr,
 812                      t_mdatoms * gmx_restrict               mdatoms,
 813                      nb_kernel_data_t * gmx_restrict        kernel_data,
 814                      t_nrnb * gmx_restrict                  nrnb)
 815 {
 816     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 817      * just 0 for non-waters.
 818      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 819      * jnr indices corresponding to data put in the four positions in the SIMD register.
 820      */
 821     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 822     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 823     int              jnrA,jnrB,jnrC,jnrD;
 824     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 825     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 826     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 827     real             rcutoff_scalar;
 828     real             *shiftvec,*fshift,*x,*f;
 829     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 830     real             scratch[4*DIM];
 831     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 832     int              vdwioffset0;
 833     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 834     int              vdwioffset1;
 835     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 836     int              vdwioffset2;
 837     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 838     int              vdwioffset3;
 839     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 840     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 841     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 842     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 843     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 844     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 845     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 846     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 847     real             *charge;
 848     int              nvdwtype;
 849     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 850     int              *vdwtype;
 851     real             *vdwparam;
 852     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 853     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 854     __m128i          ewitab;
 855     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 856     real             *ewtab;
 857     __m128           dummy_mask,cutoff_mask;
 858     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 859     __m128           one     = _mm_set1_ps(1.0);
 860     __m128           two     = _mm_set1_ps(2.0);
 861     x                = xx[0];
 862     f                = ff[0];
 863
 864     nri              = nlist->nri;
 865     iinr             = nlist->iinr;
 866     jindex           = nlist->jindex;
 867     jjnr             = nlist->jjnr;
 868     shiftidx         = nlist->shift;
 869     gid              = nlist->gid;
 870     shiftvec         = fr->shift_vec[0];
 871     fshift           = fr->fshift[0];
 872     facel            = _mm_set1_ps(fr->epsfac);
 873     charge           = mdatoms->chargeA;
 874     nvdwtype         = fr->ntype;
 875     vdwparam         = fr->nbfp;
 876     vdwtype          = mdatoms->typeA;
 877
 878     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 879     ewtab            = fr->ic->tabq_coul_F;
 880     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 881     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 882
 883     /* Setup water-specific parameters */
 884     inr              = nlist->iinr[0];
 885     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 886     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 887     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 888     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 889
 890     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 891     rcutoff_scalar   = fr->rcoulomb;
 892     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 893     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 894
 895     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 896     rvdw             = _mm_set1_ps(fr->rvdw);
 897
 898     /* Avoid stupid compiler warnings */
 899     jnrA = jnrB = jnrC = jnrD = 0;
 900     j_coord_offsetA = 0;
 901     j_coord_offsetB = 0;
 902     j_coord_offsetC = 0;
 903     j_coord_offsetD = 0;
 904
 905     outeriter        = 0;
 906     inneriter        = 0;
 907
 908     for(iidx=0;iidx<4*DIM;iidx++)
 909     {
 910         scratch[iidx] = 0.0;
 911     }
 912
 913     /* Start outer loop over neighborlists */
 914     for(iidx=0; iidx<nri; iidx++)
 915     {
 916         /* Load shift vector for this list */
 917         i_shift_offset   = DIM*shiftidx[iidx];
 918
 919         /* Load limits for loop over neighbors */
 920         j_index_start    = jindex[iidx];
 921         j_index_end      = jindex[iidx+1];
 922
 923         /* Get outer coordinate index */
 924         inr              = iinr[iidx];
 925         i_coord_offset   = DIM*inr;
 926
 927         /* Load i particle coords and add shift vector */
 928         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 929                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 930
 931         fix0             = _mm_setzero_ps();
 932         fiy0             = _mm_setzero_ps();
 933         fiz0             = _mm_setzero_ps();
 934         fix1             = _mm_setzero_ps();
 935         fiy1             = _mm_setzero_ps();
 936         fiz1             = _mm_setzero_ps();
 937         fix2             = _mm_setzero_ps();
 938         fiy2             = _mm_setzero_ps();
 939         fiz2             = _mm_setzero_ps();
 940         fix3             = _mm_setzero_ps();
 941         fiy3             = _mm_setzero_ps();
 942         fiz3             = _mm_setzero_ps();
 943
 944         /* Start inner kernel loop */
 945         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 946         {
 947
 948             /* Get j neighbor index, and coordinate index */
 949             jnrA             = jjnr[jidx];
 950             jnrB             = jjnr[jidx+1];
 951             jnrC             = jjnr[jidx+2];
 952             jnrD             = jjnr[jidx+3];
 953             j_coord_offsetA  = DIM*jnrA;
 954             j_coord_offsetB  = DIM*jnrB;
 955             j_coord_offsetC  = DIM*jnrC;
 956             j_coord_offsetD  = DIM*jnrD;
 957
 958             /* load j atom coordinates */
 959             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 960                                               x+j_coord_offsetC,x+j_coord_offsetD,
 961                                               &jx0,&jy0,&jz0);
 962
 963             /* Calculate displacement vector */
 964             dx00             = _mm_sub_ps(ix0,jx0);
 965             dy00             = _mm_sub_ps(iy0,jy0);
 966             dz00             = _mm_sub_ps(iz0,jz0);
 967             dx10             = _mm_sub_ps(ix1,jx0);
 968             dy10             = _mm_sub_ps(iy1,jy0);
 969             dz10             = _mm_sub_ps(iz1,jz0);
 970             dx20             = _mm_sub_ps(ix2,jx0);
 971             dy20             = _mm_sub_ps(iy2,jy0);
 972             dz20             = _mm_sub_ps(iz2,jz0);
 973             dx30             = _mm_sub_ps(ix3,jx0);
 974             dy30             = _mm_sub_ps(iy3,jy0);
 975             dz30             = _mm_sub_ps(iz3,jz0);
 976
 977             /* Calculate squared distance and things based on it */
 978             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 979             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 980             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 981             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 982
 983             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 984             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 985             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 986
 987             rinvsq00         = gmx_mm_inv_ps(rsq00);
 988             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 989             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 990             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 991
 992             /* Load parameters for j particles */
 993             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 994                                                               charge+jnrC+0,charge+jnrD+0);
 995             vdwjidx0A        = 2*vdwtype[jnrA+0];
 996             vdwjidx0B        = 2*vdwtype[jnrB+0];
 997             vdwjidx0C        = 2*vdwtype[jnrC+0];
 998             vdwjidx0D        = 2*vdwtype[jnrD+0];
 999
1000             fjx0             = _mm_setzero_ps();
1001             fjy0             = _mm_setzero_ps();
1002             fjz0             = _mm_setzero_ps();
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             if (gmx_mm_any_lt(rsq00,rcutoff2))
1009             {
1010
1011             /* Compute parameters for interactions between i and j atoms */
1012             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1013                                          vdwparam+vdwioffset0+vdwjidx0B,
1014                                          vdwparam+vdwioffset0+vdwjidx0C,
1015                                          vdwparam+vdwioffset0+vdwjidx0D,
1016                                          &c6_00,&c12_00);
1017
1018             /* LENNARD-JONES DISPERSION/REPULSION */
1019
1020             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1021             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1022
1023             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1024
1025             fscal            = fvdw;
1026
1027             fscal            = _mm_and_ps(fscal,cutoff_mask);
1028
1029             /* Calculate temporary vectorial force */
1030             tx               = _mm_mul_ps(fscal,dx00);
1031             ty               = _mm_mul_ps(fscal,dy00);
1032             tz               = _mm_mul_ps(fscal,dz00);
1033
1034             /* Update vectorial force */
1035             fix0             = _mm_add_ps(fix0,tx);
1036             fiy0             = _mm_add_ps(fiy0,ty);
1037             fiz0             = _mm_add_ps(fiz0,tz);
1038
1039             fjx0             = _mm_add_ps(fjx0,tx);
1040             fjy0             = _mm_add_ps(fjy0,ty);
1041             fjz0             = _mm_add_ps(fjz0,tz);
1042
1043             }
1044
1045             /**************************
1046              * CALCULATE INTERACTIONS *
1047              **************************/
1048
1049             if (gmx_mm_any_lt(rsq10,rcutoff2))
1050             {
1051
1052             r10              = _mm_mul_ps(rsq10,rinv10);
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             qq10             = _mm_mul_ps(iq1,jq0);
1056
1057             /* EWALD ELECTROSTATICS */
1058
1059             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1060             ewrt             = _mm_mul_ps(r10,ewtabscale);
1061             ewitab           = _mm_cvttps_epi32(ewrt);
1062             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1063             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1064                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1065                                          &ewtabF,&ewtabFn);
1066             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1067             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1068
1069             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1070
1071             fscal            = felec;
1072
1073             fscal            = _mm_and_ps(fscal,cutoff_mask);
1074
1075             /* Calculate temporary vectorial force */
1076             tx               = _mm_mul_ps(fscal,dx10);
1077             ty               = _mm_mul_ps(fscal,dy10);
1078             tz               = _mm_mul_ps(fscal,dz10);
1079
1080             /* Update vectorial force */
1081             fix1             = _mm_add_ps(fix1,tx);
1082             fiy1             = _mm_add_ps(fiy1,ty);
1083             fiz1             = _mm_add_ps(fiz1,tz);
1084
1085             fjx0             = _mm_add_ps(fjx0,tx);
1086             fjy0             = _mm_add_ps(fjy0,ty);
1087             fjz0             = _mm_add_ps(fjz0,tz);
1088
1089             }
1090
1091             /**************************
1092              * CALCULATE INTERACTIONS *
1093              **************************/
1094
1095             if (gmx_mm_any_lt(rsq20,rcutoff2))
1096             {
1097
1098             r20              = _mm_mul_ps(rsq20,rinv20);
1099
1100             /* Compute parameters for interactions between i and j atoms */
1101             qq20             = _mm_mul_ps(iq2,jq0);
1102
1103             /* EWALD ELECTROSTATICS */
1104
1105             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1106             ewrt             = _mm_mul_ps(r20,ewtabscale);
1107             ewitab           = _mm_cvttps_epi32(ewrt);
1108             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1109             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1110                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1111                                          &ewtabF,&ewtabFn);
1112             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1113             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1114
1115             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1116
1117             fscal            = felec;
1118
1119             fscal            = _mm_and_ps(fscal,cutoff_mask);
1120
1121             /* Calculate temporary vectorial force */
1122             tx               = _mm_mul_ps(fscal,dx20);
1123             ty               = _mm_mul_ps(fscal,dy20);
1124             tz               = _mm_mul_ps(fscal,dz20);
1125
1126             /* Update vectorial force */
1127             fix2             = _mm_add_ps(fix2,tx);
1128             fiy2             = _mm_add_ps(fiy2,ty);
1129             fiz2             = _mm_add_ps(fiz2,tz);
1130
1131             fjx0             = _mm_add_ps(fjx0,tx);
1132             fjy0             = _mm_add_ps(fjy0,ty);
1133             fjz0             = _mm_add_ps(fjz0,tz);
1134
1135             }
1136
1137             /**************************
1138              * CALCULATE INTERACTIONS *
1139              **************************/
1140
1141             if (gmx_mm_any_lt(rsq30,rcutoff2))
1142             {
1143
1144             r30              = _mm_mul_ps(rsq30,rinv30);
1145
1146             /* Compute parameters for interactions between i and j atoms */
1147             qq30             = _mm_mul_ps(iq3,jq0);
1148
1149             /* EWALD ELECTROSTATICS */
1150
1151             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1152             ewrt             = _mm_mul_ps(r30,ewtabscale);
1153             ewitab           = _mm_cvttps_epi32(ewrt);
1154             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1155             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1156                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1157                                          &ewtabF,&ewtabFn);
1158             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1159             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1160
1161             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm_and_ps(fscal,cutoff_mask);
1166
1167             /* Calculate temporary vectorial force */
1168             tx               = _mm_mul_ps(fscal,dx30);
1169             ty               = _mm_mul_ps(fscal,dy30);
1170             tz               = _mm_mul_ps(fscal,dz30);
1171
1172             /* Update vectorial force */
1173             fix3             = _mm_add_ps(fix3,tx);
1174             fiy3             = _mm_add_ps(fiy3,ty);
1175             fiz3             = _mm_add_ps(fiz3,tz);
1176
1177             fjx0             = _mm_add_ps(fjx0,tx);
1178             fjy0             = _mm_add_ps(fjy0,ty);
1179             fjz0             = _mm_add_ps(fjz0,tz);
1180
1181             }
1182
1183             fjptrA             = f+j_coord_offsetA;
1184             fjptrB             = f+j_coord_offsetB;
1185             fjptrC             = f+j_coord_offsetC;
1186             fjptrD             = f+j_coord_offsetD;
1187
1188             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1189
1190             /* Inner loop uses 147 flops */
1191         }
1192
1193         if(jidx<j_index_end)
1194         {
1195
1196             /* Get j neighbor index, and coordinate index */
1197             jnrlistA         = jjnr[jidx];
1198             jnrlistB         = jjnr[jidx+1];
1199             jnrlistC         = jjnr[jidx+2];
1200             jnrlistD         = jjnr[jidx+3];
1201             /* Sign of each element will be negative for non-real atoms.
1202              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1203              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1204              */
1205             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1206             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1207             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1208             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1209             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1210             j_coord_offsetA  = DIM*jnrA;
1211             j_coord_offsetB  = DIM*jnrB;
1212             j_coord_offsetC  = DIM*jnrC;
1213             j_coord_offsetD  = DIM*jnrD;
1214
1215             /* load j atom coordinates */
1216             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1217                                               x+j_coord_offsetC,x+j_coord_offsetD,
1218                                               &jx0,&jy0,&jz0);
1219
1220             /* Calculate displacement vector */
1221             dx00             = _mm_sub_ps(ix0,jx0);
1222             dy00             = _mm_sub_ps(iy0,jy0);
1223             dz00             = _mm_sub_ps(iz0,jz0);
1224             dx10             = _mm_sub_ps(ix1,jx0);
1225             dy10             = _mm_sub_ps(iy1,jy0);
1226             dz10             = _mm_sub_ps(iz1,jz0);
1227             dx20             = _mm_sub_ps(ix2,jx0);
1228             dy20             = _mm_sub_ps(iy2,jy0);
1229             dz20             = _mm_sub_ps(iz2,jz0);
1230             dx30             = _mm_sub_ps(ix3,jx0);
1231             dy30             = _mm_sub_ps(iy3,jy0);
1232             dz30             = _mm_sub_ps(iz3,jz0);
1233
1234             /* Calculate squared distance and things based on it */
1235             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1236             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1237             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1238             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1239
1240             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1241             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1242             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1243
1244             rinvsq00         = gmx_mm_inv_ps(rsq00);
1245             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1246             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1247             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1248
1249             /* Load parameters for j particles */
1250             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1251                                                               charge+jnrC+0,charge+jnrD+0);
1252             vdwjidx0A        = 2*vdwtype[jnrA+0];
1253             vdwjidx0B        = 2*vdwtype[jnrB+0];
1254             vdwjidx0C        = 2*vdwtype[jnrC+0];
1255             vdwjidx0D        = 2*vdwtype[jnrD+0];
1256
1257             fjx0             = _mm_setzero_ps();
1258             fjy0             = _mm_setzero_ps();
1259             fjz0             = _mm_setzero_ps();
1260
1261             /**************************
1262              * CALCULATE INTERACTIONS *
1263              **************************/
1264
1265             if (gmx_mm_any_lt(rsq00,rcutoff2))
1266             {
1267
1268             /* Compute parameters for interactions between i and j atoms */
1269             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1270                                          vdwparam+vdwioffset0+vdwjidx0B,
1271                                          vdwparam+vdwioffset0+vdwjidx0C,
1272                                          vdwparam+vdwioffset0+vdwjidx0D,
1273                                          &c6_00,&c12_00);
1274
1275             /* LENNARD-JONES DISPERSION/REPULSION */
1276
1277             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1278             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1279
1280             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1281
1282             fscal            = fvdw;
1283
1284             fscal            = _mm_and_ps(fscal,cutoff_mask);
1285
1286             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1287
1288             /* Calculate temporary vectorial force */
1289             tx               = _mm_mul_ps(fscal,dx00);
1290             ty               = _mm_mul_ps(fscal,dy00);
1291             tz               = _mm_mul_ps(fscal,dz00);
1292
1293             /* Update vectorial force */
1294             fix0             = _mm_add_ps(fix0,tx);
1295             fiy0             = _mm_add_ps(fiy0,ty);
1296             fiz0             = _mm_add_ps(fiz0,tz);
1297
1298             fjx0             = _mm_add_ps(fjx0,tx);
1299             fjy0             = _mm_add_ps(fjy0,ty);
1300             fjz0             = _mm_add_ps(fjz0,tz);
1301
1302             }
1303
1304             /**************************
1305              * CALCULATE INTERACTIONS *
1306              **************************/
1307
1308             if (gmx_mm_any_lt(rsq10,rcutoff2))
1309             {
1310
1311             r10              = _mm_mul_ps(rsq10,rinv10);
1312             r10              = _mm_andnot_ps(dummy_mask,r10);
1313
1314             /* Compute parameters for interactions between i and j atoms */
1315             qq10             = _mm_mul_ps(iq1,jq0);
1316
1317             /* EWALD ELECTROSTATICS */
1318
1319             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1320             ewrt             = _mm_mul_ps(r10,ewtabscale);
1321             ewitab           = _mm_cvttps_epi32(ewrt);
1322             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1323             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1324                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1325                                          &ewtabF,&ewtabFn);
1326             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1327             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1328
1329             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1330
1331             fscal            = felec;
1332
1333             fscal            = _mm_and_ps(fscal,cutoff_mask);
1334
1335             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1336
1337             /* Calculate temporary vectorial force */
1338             tx               = _mm_mul_ps(fscal,dx10);
1339             ty               = _mm_mul_ps(fscal,dy10);
1340             tz               = _mm_mul_ps(fscal,dz10);
1341
1342             /* Update vectorial force */
1343             fix1             = _mm_add_ps(fix1,tx);
1344             fiy1             = _mm_add_ps(fiy1,ty);
1345             fiz1             = _mm_add_ps(fiz1,tz);
1346
1347             fjx0             = _mm_add_ps(fjx0,tx);
1348             fjy0             = _mm_add_ps(fjy0,ty);
1349             fjz0             = _mm_add_ps(fjz0,tz);
1350
1351             }
1352
1353             /**************************
1354              * CALCULATE INTERACTIONS *
1355              **************************/
1356
1357             if (gmx_mm_any_lt(rsq20,rcutoff2))
1358             {
1359
1360             r20              = _mm_mul_ps(rsq20,rinv20);
1361             r20              = _mm_andnot_ps(dummy_mask,r20);
1362
1363             /* Compute parameters for interactions between i and j atoms */
1364             qq20             = _mm_mul_ps(iq2,jq0);
1365
1366             /* EWALD ELECTROSTATICS */
1367
1368             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1369             ewrt             = _mm_mul_ps(r20,ewtabscale);
1370             ewitab           = _mm_cvttps_epi32(ewrt);
1371             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1372             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1373                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1374                                          &ewtabF,&ewtabFn);
1375             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1376             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1377
1378             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1379
1380             fscal            = felec;
1381
1382             fscal            = _mm_and_ps(fscal,cutoff_mask);
1383
1384             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1385
1386             /* Calculate temporary vectorial force */
1387             tx               = _mm_mul_ps(fscal,dx20);
1388             ty               = _mm_mul_ps(fscal,dy20);
1389             tz               = _mm_mul_ps(fscal,dz20);
1390
1391             /* Update vectorial force */
1392             fix2             = _mm_add_ps(fix2,tx);
1393             fiy2             = _mm_add_ps(fiy2,ty);
1394             fiz2             = _mm_add_ps(fiz2,tz);
1395
1396             fjx0             = _mm_add_ps(fjx0,tx);
1397             fjy0             = _mm_add_ps(fjy0,ty);
1398             fjz0             = _mm_add_ps(fjz0,tz);
1399
1400             }
1401
1402             /**************************
1403              * CALCULATE INTERACTIONS *
1404              **************************/
1405
1406             if (gmx_mm_any_lt(rsq30,rcutoff2))
1407             {
1408
1409             r30              = _mm_mul_ps(rsq30,rinv30);
1410             r30              = _mm_andnot_ps(dummy_mask,r30);
1411
1412             /* Compute parameters for interactions between i and j atoms */
1413             qq30             = _mm_mul_ps(iq3,jq0);
1414
1415             /* EWALD ELECTROSTATICS */
1416
1417             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1418             ewrt             = _mm_mul_ps(r30,ewtabscale);
1419             ewitab           = _mm_cvttps_epi32(ewrt);
1420             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1421             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1422                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1423                                          &ewtabF,&ewtabFn);
1424             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1425             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1426
1427             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1428
1429             fscal            = felec;
1430
1431             fscal            = _mm_and_ps(fscal,cutoff_mask);
1432
1433             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1434
1435             /* Calculate temporary vectorial force */
1436             tx               = _mm_mul_ps(fscal,dx30);
1437             ty               = _mm_mul_ps(fscal,dy30);
1438             tz               = _mm_mul_ps(fscal,dz30);
1439
1440             /* Update vectorial force */
1441             fix3             = _mm_add_ps(fix3,tx);
1442             fiy3             = _mm_add_ps(fiy3,ty);
1443             fiz3             = _mm_add_ps(fiz3,tz);
1444
1445             fjx0             = _mm_add_ps(fjx0,tx);
1446             fjy0             = _mm_add_ps(fjy0,ty);
1447             fjz0             = _mm_add_ps(fjz0,tz);
1448
1449             }
1450
1451             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1452             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1453             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1454             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1455
1456             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1457
1458             /* Inner loop uses 150 flops */
1459         }
1460
1461         /* End of innermost loop */
1462
1463         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1464                                               f+i_coord_offset,fshift+i_shift_offset);
1465
1466         /* Increment number of inner iterations */
1467         inneriter                  += j_index_end - j_index_start;
1468
1469         /* Outer loop uses 24 flops */
1470     }
1471
1472     /* Increment number of outer iterations */
1473     outeriter        += nri;
1474
1475     /* Update outer/inner flops */
1476
1477     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1478 }